diff options
author | dim <dim@FreeBSD.org> | 2014-03-21 17:53:59 +0000 |
---|---|---|
committer | dim <dim@FreeBSD.org> | 2014-03-21 17:53:59 +0000 |
commit | 9cedb8bb69b89b0f0c529937247a6a80cabdbaec (patch) | |
tree | c978f0e9ec1ab92dc8123783f30b08a7fd1e2a39 /contrib/llvm/lib/Transforms/IPO | |
parent | 03fdc2934eb61c44c049a02b02aa974cfdd8a0eb (diff) | |
download | FreeBSD-src-9cedb8bb69b89b0f0c529937247a6a80cabdbaec.zip FreeBSD-src-9cedb8bb69b89b0f0c529937247a6a80cabdbaec.tar.gz |
MFC 261991:
Upgrade our copy of llvm/clang to 3.4 release. This version supports
all of the features in the current working draft of the upcoming C++
standard, provisionally named C++1y.
The code generator's performance is greatly increased, and the loop
auto-vectorizer is now enabled at -Os and -O2 in addition to -O3. The
PowerPC backend has made several major improvements to code generation
quality and compile time, and the X86, SPARC, ARM32, Aarch64 and SystemZ
backends have all seen major feature work.
Release notes for llvm and clang can be found here:
<http://llvm.org/releases/3.4/docs/ReleaseNotes.html>
<http://llvm.org/releases/3.4/tools/clang/docs/ReleaseNotes.html>
MFC 262121 (by emaste):
Update lldb for clang/llvm 3.4 import
This commit largely restores the lldb source to the upstream r196259
snapshot with the addition of threaded inferior support and a few bug
fixes.
Specific upstream lldb revisions restored include:
SVN git
181387 779e6ac
181703 7bef4e2
182099 b31044e
182650 f2dcf35
182683 0d91b80
183862 15c1774
183929 99447a6
184177 0b2934b
184948 4dc3761
184954 007e7bc
186990 eebd175
Sponsored by: DARPA, AFRL
MFC 262186 (by emaste):
Fix mismerge in r262121
A break statement was lost in the merge. The error had no functional
impact, but restore it to reduce the diff against upstream.
MFC 262303:
Pull in r197521 from upstream clang trunk (by rdivacky):
Use the integrated assembler by default on FreeBSD/ppc and ppc64.
Requested by: jhibbits
MFC 262611:
Pull in r196874 from upstream llvm trunk:
Fix a crash that occurs when PWD is invalid.
MCJIT needs to be able to run in hostile environments, even when PWD
is invalid. There's no need to crash MCJIT in this case.
The obvious fix is to simply leave MCContext's CompilationDir empty
when PWD can't be determined. This way, MCJIT clients,
and other clients that link with LLVM don't need a valid working directory.
If we do want to guarantee valid CompilationDir, that should be done
only for clients of getCompilationDir(). This is as simple as checking
for an empty string.
The only current use of getCompilationDir is EmitGenDwarfInfo, which
won't conceivably run with an invalid working dir. However, in the
purely hypothetically and untestable case that this happens, the
AT_comp_dir will be omitted from the compilation_unit DIE.
This should help fix assertions occurring with ports-mgmt/tinderbox,
when it is using jails, and sometimes invalidates clang's current
working directory.
Reported by: decke
MFC 262809:
Pull in r203007 from upstream clang trunk:
Don't produce an alias between destructors with different calling conventions.
Fixes pr19007.
(Please note that is an LLVM PR identifier, not a FreeBSD one.)
This should fix Firefox and/or libxul crashes (due to problems with
regparm/stdcall calling conventions) on i386.
Reported by: multiple users on freebsd-current
PR: bin/187103
MFC 263048:
Repair recognition of "CC" as an alias for the C++ compiler, since it
was silently broken by upstream for a Windows-specific use-case.
Apparently some versions of CMake still rely on this archaic feature...
Reported by: rakuco
MFC 263049:
Garbage collect the old way of adding the libstdc++ include directories
in clang's InitHeaderSearch.cpp. This has been superseded by David
Chisnall's commit in r255321.
Moreover, if libc++ is used, the libstdc++ include directories should
not be in the search path at all. These directories are now only used
if you pass -stdlib=libstdc++.
Diffstat (limited to 'contrib/llvm/lib/Transforms/IPO')
-rw-r--r-- | contrib/llvm/lib/Transforms/IPO/ArgumentPromotion.cpp | 31 | ||||
-rw-r--r-- | contrib/llvm/lib/Transforms/IPO/ConstantMerge.cpp | 11 | ||||
-rw-r--r-- | contrib/llvm/lib/Transforms/IPO/DeadArgumentElimination.cpp | 138 | ||||
-rw-r--r-- | contrib/llvm/lib/Transforms/IPO/ExtractGV.cpp | 54 | ||||
-rw-r--r-- | contrib/llvm/lib/Transforms/IPO/FunctionAttrs.cpp | 437 | ||||
-rw-r--r-- | contrib/llvm/lib/Transforms/IPO/GlobalDCE.cpp | 3 | ||||
-rw-r--r-- | contrib/llvm/lib/Transforms/IPO/GlobalOpt.cpp | 514 | ||||
-rw-r--r-- | contrib/llvm/lib/Transforms/IPO/InlineAlways.cpp | 2 | ||||
-rw-r--r-- | contrib/llvm/lib/Transforms/IPO/InlineSimple.cpp | 4 | ||||
-rw-r--r-- | contrib/llvm/lib/Transforms/IPO/Inliner.cpp | 27 | ||||
-rw-r--r-- | contrib/llvm/lib/Transforms/IPO/Internalize.cpp | 136 | ||||
-rw-r--r-- | contrib/llvm/lib/Transforms/IPO/MergeFunctions.cpp | 77 | ||||
-rw-r--r-- | contrib/llvm/lib/Transforms/IPO/PassManagerBuilder.cpp | 62 | ||||
-rw-r--r-- | contrib/llvm/lib/Transforms/IPO/PruneEH.cpp | 14 | ||||
-rw-r--r-- | contrib/llvm/lib/Transforms/IPO/StripSymbols.cpp | 231 |
15 files changed, 1045 insertions, 696 deletions
diff --git a/contrib/llvm/lib/Transforms/IPO/ArgumentPromotion.cpp b/contrib/llvm/lib/Transforms/IPO/ArgumentPromotion.cpp index e6fa4ed..df08091 100644 --- a/contrib/llvm/lib/Transforms/IPO/ArgumentPromotion.cpp +++ b/contrib/llvm/lib/Transforms/IPO/ArgumentPromotion.cpp @@ -88,7 +88,7 @@ char ArgPromotion::ID = 0; INITIALIZE_PASS_BEGIN(ArgPromotion, "argpromotion", "Promote 'by reference' arguments to scalars", false, false) INITIALIZE_AG_DEPENDENCY(AliasAnalysis) -INITIALIZE_AG_DEPENDENCY(CallGraph) +INITIALIZE_PASS_DEPENDENCY(CallGraph) INITIALIZE_PASS_END(ArgPromotion, "argpromotion", "Promote 'by reference' arguments to scalars", false, false) @@ -126,12 +126,10 @@ CallGraphNode *ArgPromotion::PromoteArguments(CallGraphNode *CGN) { if (!F || !F->hasLocalLinkage()) return 0; // First check: see if there are any pointer arguments! If not, quick exit. - SmallVector<std::pair<Argument*, unsigned>, 16> PointerArgs; - unsigned ArgNo = 0; - for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end(); - I != E; ++I, ++ArgNo) + SmallVector<Argument*, 16> PointerArgs; + for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end(); I != E; ++I) if (I->getType()->isPointerTy()) - PointerArgs.push_back(std::pair<Argument*, unsigned>(I, ArgNo)); + PointerArgs.push_back(I); if (PointerArgs.empty()) return 0; // Second check: make sure that all callers are direct callers. We can't @@ -152,15 +150,13 @@ CallGraphNode *ArgPromotion::PromoteArguments(CallGraphNode *CGN) { // add it to ArgsToPromote. SmallPtrSet<Argument*, 8> ArgsToPromote; SmallPtrSet<Argument*, 8> ByValArgsToTransform; - for (unsigned i = 0; i != PointerArgs.size(); ++i) { - bool isByVal=F->getAttributes(). - hasAttribute(PointerArgs[i].second+1, Attribute::ByVal); - Argument *PtrArg = PointerArgs[i].first; + for (unsigned i = 0, e = PointerArgs.size(); i != e; ++i) { + Argument *PtrArg = PointerArgs[i]; Type *AgTy = cast<PointerType>(PtrArg->getType())->getElementType(); // If this is a byval argument, and if the aggregate type is small, just // pass the elements, which is always safe. - if (isByVal) { + if (PtrArg->hasByValAttr()) { if (StructType *STy = dyn_cast<StructType>(AgTy)) { if (maxElements > 0 && STy->getNumElements() > maxElements) { DEBUG(dbgs() << "argpromotion disable promoting argument '" @@ -205,7 +201,7 @@ CallGraphNode *ArgPromotion::PromoteArguments(CallGraphNode *CGN) { } // Otherwise, see if we can promote the pointer to its value. - if (isSafeToPromoteArgument(PtrArg, isByVal)) + if (isSafeToPromoteArgument(PtrArg, PtrArg->hasByValAttr())) ArgsToPromote.insert(PtrArg); } @@ -221,8 +217,7 @@ CallGraphNode *ArgPromotion::PromoteArguments(CallGraphNode *CGN) { static bool AllCallersPassInValidPointerForArgument(Argument *Arg) { Function *Callee = Arg->getParent(); - unsigned ArgNo = std::distance(Callee->arg_begin(), - Function::arg_iterator(Arg)); + unsigned ArgNo = Arg->getArgNo(); // Look at all call sites of the function. At this pointer we know we only // have direct callees. @@ -509,7 +504,9 @@ CallGraphNode *ArgPromotion::DoPromotion(Function *F, // OriginalLoads - Keep track of a representative load instruction from the // original function so that we can tell the alias analysis implementation // what the new GEP/Load instructions we are inserting look like. - std::map<IndicesVector, LoadInst*> OriginalLoads; + // We need to keep the original loads for each argument and the elements + // of the argument that are accessed. + std::map<std::pair<Argument*, IndicesVector>, LoadInst*> OriginalLoads; // Attribute - Keep track of the parameter attributes for the arguments // that we are *not* promoting. For the ones that we do promote, the parameter @@ -574,7 +571,7 @@ CallGraphNode *ArgPromotion::DoPromotion(Function *F, else // Take any load, we will use it only to update Alias Analysis OrigLoad = cast<LoadInst>(User->use_back()); - OriginalLoads[Indices] = OrigLoad; + OriginalLoads[std::make_pair(I, Indices)] = OrigLoad; } // Add a parameter to the function for each element passed in. @@ -681,7 +678,7 @@ CallGraphNode *ArgPromotion::DoPromotion(Function *F, for (ScalarizeTable::iterator SI = ArgIndices.begin(), E = ArgIndices.end(); SI != E; ++SI) { Value *V = *AI; - LoadInst *OrigLoad = OriginalLoads[*SI]; + LoadInst *OrigLoad = OriginalLoads[std::make_pair(I, *SI)]; if (!SI->empty()) { Ops.reserve(SI->size()); Type *ElTy = V->getType(); diff --git a/contrib/llvm/lib/Transforms/IPO/ConstantMerge.cpp b/contrib/llvm/lib/Transforms/IPO/ConstantMerge.cpp index a7bf188..d94c0f4 100644 --- a/contrib/llvm/lib/Transforms/IPO/ConstantMerge.cpp +++ b/contrib/llvm/lib/Transforms/IPO/ConstantMerge.cpp @@ -93,9 +93,12 @@ bool ConstantMerge::hasKnownAlignment(GlobalVariable *GV) const { } unsigned ConstantMerge::getAlignment(GlobalVariable *GV) const { + unsigned Align = GV->getAlignment(); + if (Align) + return Align; if (TD) return TD->getPreferredAlignment(GV); - return GV->getAlignment(); + return 0; } bool ConstantMerge::runOnModule(Module &M) { @@ -210,9 +213,9 @@ bool ConstantMerge::runOnModule(Module &M) { // Bump the alignment if necessary. if (Replacements[i].first->getAlignment() || Replacements[i].second->getAlignment()) { - Replacements[i].second->setAlignment(std::max( - Replacements[i].first->getAlignment(), - Replacements[i].second->getAlignment())); + Replacements[i].second->setAlignment( + std::max(getAlignment(Replacements[i].first), + getAlignment(Replacements[i].second))); } // Eliminate any uses of the dead global. diff --git a/contrib/llvm/lib/Transforms/IPO/DeadArgumentElimination.cpp b/contrib/llvm/lib/Transforms/IPO/DeadArgumentElimination.cpp index 49ef1e7..911c14e 100644 --- a/contrib/llvm/lib/Transforms/IPO/DeadArgumentElimination.cpp +++ b/contrib/llvm/lib/Transforms/IPO/DeadArgumentElimination.cpp @@ -211,7 +211,9 @@ void DAE::CollectFunctionDIs(Module &M) { for (unsigned SPIndex = 0, SPNum = SPs.getNumElements(); SPIndex < SPNum; ++SPIndex) { DISubprogram SP(SPs.getElement(SPIndex)); - if (!SP.Verify()) + assert((!SP || SP.isSubprogram()) && + "A MDNode in subprograms of a CU should be null or a DISubprogram."); + if (!SP) continue; if (Function *F = SP.getFunction()) FunctionDIs[F] = SP; @@ -263,8 +265,10 @@ bool DAE::DeleteDeadVarargs(Function &Fn) { // to pass in a smaller number of arguments into the new function. // std::vector<Value*> Args; - while (!Fn.use_empty()) { - CallSite CS(Fn.use_back()); + for (Value::use_iterator I = Fn.use_begin(), E = Fn.use_end(); I != E; ) { + CallSite CS(*I++); + if (!CS) + continue; Instruction *Call = CS.getInstruction(); // Pass all the same arguments. @@ -330,6 +334,11 @@ bool DAE::DeleteDeadVarargs(Function &Fn) { if (DI != FunctionDIs.end()) DI->second.replaceFunction(NF); + // Fix up any BlockAddresses that refer to the function. + Fn.replaceAllUsesWith(ConstantExpr::getBitCast(NF, Fn.getType())); + // Delete the bitcast that we just created, so that NF does not + // appear to be address-taken. + NF->removeDeadConstantUsers(); // Finally, nuke the old function. Fn.eraseFromParent(); return true; @@ -343,8 +352,22 @@ bool DAE::RemoveDeadArgumentsFromCallers(Function &Fn) if (Fn.isDeclaration() || Fn.mayBeOverridden()) return false; - // Functions with local linkage should already have been handled. - if (Fn.hasLocalLinkage()) + // Functions with local linkage should already have been handled, except the + // fragile (variadic) ones which we can improve here. + if (Fn.hasLocalLinkage() && !Fn.getFunctionType()->isVarArg()) + return false; + + // If a function seen at compile time is not necessarily the one linked to + // the binary being built, it is illegal to change the actual arguments + // passed to it. These functions can be captured by isWeakForLinker(). + // *NOTE* that mayBeOverridden() is insufficient for this purpose as it + // doesn't include linkage types like AvailableExternallyLinkage and + // LinkOnceODRLinkage. Take link_odr* as an example, it indicates a set of + // *EQUIVALENT* globals that can be merged at link-time. However, the + // semantic of *EQUIVALENT*-functions includes parameters. Changing + // parameters breaks this assumption. + // + if (Fn.isWeakForLinker()) return false; if (Fn.use_empty()) @@ -604,9 +627,20 @@ void DAE::SurveyFunction(const Function &F) { UseVector MaybeLiveArgUses; for (Function::const_arg_iterator AI = F.arg_begin(), E = F.arg_end(); AI != E; ++AI, ++i) { - // See what the effect of this use is (recording any uses that cause - // MaybeLive in MaybeLiveArgUses). - Liveness Result = SurveyUses(AI, MaybeLiveArgUses); + Liveness Result; + if (F.getFunctionType()->isVarArg()) { + // Variadic functions will already have a va_arg function expanded inside + // them, making them potentially very sensitive to ABI changes resulting + // from removing arguments entirely, so don't. For example AArch64 handles + // register and stack HFAs very differently, and this is reflected in the + // IR which has already been generated. + Result = Live; + } else { + // See what the effect of this use is (recording any uses that cause + // MaybeLive in MaybeLiveArgUses). + Result = SurveyUses(AI, MaybeLiveArgUses); + } + // Mark the result. MarkValue(CreateArg(&F, i), Result, MaybeLiveArgUses); // Clear the vector again for the next iteration. @@ -695,10 +729,42 @@ bool DAE::RemoveDeadStuffFromFunction(Function *F) { FunctionType *FTy = F->getFunctionType(); std::vector<Type*> Params; + // Keep track of if we have a live 'returned' argument + bool HasLiveReturnedArg = false; + // Set up to build a new list of parameter attributes. SmallVector<AttributeSet, 8> AttributesVec; const AttributeSet &PAL = F->getAttributes(); + // Remember which arguments are still alive. + SmallVector<bool, 10> ArgAlive(FTy->getNumParams(), false); + // Construct the new parameter list from non-dead arguments. Also construct + // a new set of parameter attributes to correspond. Skip the first parameter + // attribute, since that belongs to the return value. + unsigned i = 0; + for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end(); + I != E; ++I, ++i) { + RetOrArg Arg = CreateArg(F, i); + if (LiveValues.erase(Arg)) { + Params.push_back(I->getType()); + ArgAlive[i] = true; + + // Get the original parameter attributes (skipping the first one, that is + // for the return value. + if (PAL.hasAttributes(i + 1)) { + AttrBuilder B(PAL, i + 1); + if (B.contains(Attribute::Returned)) + HasLiveReturnedArg = true; + AttributesVec. + push_back(AttributeSet::get(F->getContext(), Params.size(), B)); + } + } else { + ++NumArgumentsEliminated; + DEBUG(dbgs() << "DAE - Removing argument " << i << " (" << I->getName() + << ") from " << F->getName() << "\n"); + } + } + // Find out the new return value. Type *RetTy = FTy->getReturnType(); Type *NRetTy = NULL; @@ -707,7 +773,27 @@ bool DAE::RemoveDeadStuffFromFunction(Function *F) { // -1 means unused, other numbers are the new index SmallVector<int, 5> NewRetIdxs(RetCount, -1); std::vector<Type*> RetTypes; - if (RetTy->isVoidTy()) { + + // If there is a function with a live 'returned' argument but a dead return + // value, then there are two possible actions: + // 1) Eliminate the return value and take off the 'returned' attribute on the + // argument. + // 2) Retain the 'returned' attribute and treat the return value (but not the + // entire function) as live so that it is not eliminated. + // + // It's not clear in the general case which option is more profitable because, + // even in the absence of explicit uses of the return value, code generation + // is free to use the 'returned' attribute to do things like eliding + // save/restores of registers across calls. Whether or not this happens is + // target and ABI-specific as well as depending on the amount of register + // pressure, so there's no good way for an IR-level pass to figure this out. + // + // Fortunately, the only places where 'returned' is currently generated by + // the FE are places where 'returned' is basically free and almost always a + // performance win, so the second option can just be used always for now. + // + // This should be revisited if 'returned' is ever applied more liberally. + if (RetTy->isVoidTy() || HasLiveReturnedArg) { NRetTy = RetTy; } else { StructType *STy = dyn_cast<StructType>(RetTy); @@ -777,33 +863,6 @@ bool DAE::RemoveDeadStuffFromFunction(Function *F) { if (RAttrs.hasAttributes(AttributeSet::ReturnIndex)) AttributesVec.push_back(AttributeSet::get(NRetTy->getContext(), RAttrs)); - // Remember which arguments are still alive. - SmallVector<bool, 10> ArgAlive(FTy->getNumParams(), false); - // Construct the new parameter list from non-dead arguments. Also construct - // a new set of parameter attributes to correspond. Skip the first parameter - // attribute, since that belongs to the return value. - unsigned i = 0; - for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end(); - I != E; ++I, ++i) { - RetOrArg Arg = CreateArg(F, i); - if (LiveValues.erase(Arg)) { - Params.push_back(I->getType()); - ArgAlive[i] = true; - - // Get the original parameter attributes (skipping the first one, that is - // for the return value. - if (PAL.hasAttributes(i + 1)) { - AttrBuilder B(PAL, i + 1); - AttributesVec. - push_back(AttributeSet::get(F->getContext(), Params.size(), B)); - } - } else { - ++NumArgumentsEliminated; - DEBUG(dbgs() << "DAE - Removing argument " << i << " (" << I->getName() - << ") from " << F->getName() << "\n"); - } - } - if (PAL.hasAttributes(AttributeSet::FunctionIndex)) AttributesVec.push_back(AttributeSet::get(F->getContext(), PAL.getFnAttributes())); @@ -864,6 +923,13 @@ bool DAE::RemoveDeadStuffFromFunction(Function *F) { // Get original parameter attributes, but skip return attributes. if (CallPAL.hasAttributes(i + 1)) { AttrBuilder B(CallPAL, i + 1); + // If the return type has changed, then get rid of 'returned' on the + // call site. The alternative is to make all 'returned' attributes on + // call sites keep the return value alive just like 'returned' + // attributes on function declaration but it's less clearly a win + // and this is not an expected case anyway + if (NRetTy != RetTy && B.contains(Attribute::Returned)) + B.removeAttribute(Attribute::Returned); AttributesVec. push_back(AttributeSet::get(F->getContext(), Args.size(), B)); } diff --git a/contrib/llvm/lib/Transforms/IPO/ExtractGV.cpp b/contrib/llvm/lib/Transforms/IPO/ExtractGV.cpp index fa3d72d..50fb3e6 100644 --- a/contrib/llvm/lib/Transforms/IPO/ExtractGV.cpp +++ b/contrib/llvm/lib/Transforms/IPO/ExtractGV.cpp @@ -21,6 +21,38 @@ #include <algorithm> using namespace llvm; +/// Make sure GV is visible from both modules. Delete is true if it is +/// being deleted from this module. +/// This also makes sure GV cannot be dropped so that references from +/// the split module remain valid. +static void makeVisible(GlobalValue &GV, bool Delete) { + bool Local = GV.hasLocalLinkage(); + if (Local) + GV.setVisibility(GlobalValue::HiddenVisibility); + + if (Local || Delete) { + GV.setLinkage(GlobalValue::ExternalLinkage); + return; + } + + if (!GV.hasLinkOnceLinkage()) { + assert(!GV.isDiscardableIfUnused()); + return; + } + + // Map linkonce* to weak* so that llvm doesn't drop this GV. + switch(GV.getLinkage()) { + default: + llvm_unreachable("Unexpected linkage"); + case GlobalValue::LinkOnceAnyLinkage: + GV.setLinkage(GlobalValue::WeakAnyLinkage); + return; + case GlobalValue::LinkOnceODRLinkage: + GV.setLinkage(GlobalValue::WeakODRLinkage); + return; + } +} + namespace { /// @brief A pass to extract specific functions and their dependencies. class GVExtractorPass : public ModulePass { @@ -60,12 +92,7 @@ namespace { continue; } - bool Local = I->isDiscardableIfUnused(); - if (Local) - I->setVisibility(GlobalValue::HiddenVisibility); - - if (Local || Delete) - I->setLinkage(GlobalValue::ExternalLinkage); + makeVisible(*I, Delete); if (Delete) I->setInitializer(0); @@ -80,12 +107,7 @@ namespace { continue; } - bool Local = I->isDiscardableIfUnused(); - if (Local) - I->setVisibility(GlobalValue::HiddenVisibility); - - if (Local || Delete) - I->setLinkage(GlobalValue::ExternalLinkage); + makeVisible(*I, Delete); if (Delete) I->deleteBody(); @@ -97,12 +119,10 @@ namespace { Module::alias_iterator CurI = I; ++I; - if (CurI->isDiscardableIfUnused()) { - CurI->setVisibility(GlobalValue::HiddenVisibility); - CurI->setLinkage(GlobalValue::ExternalLinkage); - } + bool Delete = deleteStuff == (bool)Named.count(CurI); + makeVisible(*CurI, Delete); - if (deleteStuff == (bool)Named.count(CurI)) { + if (Delete) { Type *Ty = CurI->getType()->getElementType(); CurI->removeFromParent(); diff --git a/contrib/llvm/lib/Transforms/IPO/FunctionAttrs.cpp b/contrib/llvm/lib/Transforms/IPO/FunctionAttrs.cpp index bc5109b..60e5f06 100644 --- a/contrib/llvm/lib/Transforms/IPO/FunctionAttrs.cpp +++ b/contrib/llvm/lib/Transforms/IPO/FunctionAttrs.cpp @@ -9,14 +9,12 @@ // // This file implements a simple interprocedural pass which walks the // call-graph, looking for functions which do not access or only read -// non-local memory, and marking them readnone/readonly. In addition, -// it marks function arguments (of pointer type) 'nocapture' if a call -// to the function does not create any copies of the pointer value that -// outlive the call. This more or less means that the pointer is only -// dereferenced, and not returned from the function or stored in a global. -// Finally, well-known library call declarations are marked with all -// attributes that are consistent with the function's standard definition. -// This pass is implemented as a bottom-up traversal of the call-graph. +// non-local memory, and marking them readnone/readonly. It does the +// same with function arguments independently, marking them readonly/ +// readnone/nocapture. Finally, well-known library call declarations +// are marked with all attributes that are consistent with the +// function's standard definition. This pass is implemented as a +// bottom-up traversal of the call-graph. // //===----------------------------------------------------------------------===// @@ -40,6 +38,8 @@ using namespace llvm; STATISTIC(NumReadNone, "Number of functions marked readnone"); STATISTIC(NumReadOnly, "Number of functions marked readonly"); STATISTIC(NumNoCapture, "Number of arguments marked nocapture"); +STATISTIC(NumReadNoneArg, "Number of arguments marked readnone"); +STATISTIC(NumReadOnlyArg, "Number of arguments marked readonly"); STATISTIC(NumNoAlias, "Number of function returns marked noalias"); STATISTIC(NumAnnotated, "Number of attributes added to library functions"); @@ -56,8 +56,8 @@ namespace { // AddReadAttrs - Deduce readonly/readnone attributes for the SCC. bool AddReadAttrs(const CallGraphSCC &SCC); - // AddNoCaptureAttrs - Deduce nocapture attributes for the SCC. - bool AddNoCaptureAttrs(const CallGraphSCC &SCC); + // AddArgumentAttrs - Deduce nocapture attributes for the SCC. + bool AddArgumentAttrs(const CallGraphSCC &SCC); // IsFunctionMallocLike - Does this function allocate new memory? bool IsFunctionMallocLike(Function *F, @@ -71,36 +71,43 @@ namespace { void setDoesNotAccessMemory(Function &F) { if (!F.doesNotAccessMemory()) { - F.setDoesNotAccessMemory(); - ++NumAnnotated; + F.setDoesNotAccessMemory(); + ++NumAnnotated; } } void setOnlyReadsMemory(Function &F) { if (!F.onlyReadsMemory()) { - F.setOnlyReadsMemory(); - ++NumAnnotated; + F.setOnlyReadsMemory(); + ++NumAnnotated; } } void setDoesNotThrow(Function &F) { if (!F.doesNotThrow()) { - F.setDoesNotThrow(); - ++NumAnnotated; + F.setDoesNotThrow(); + ++NumAnnotated; } } void setDoesNotCapture(Function &F, unsigned n) { if (!F.doesNotCapture(n)) { - F.setDoesNotCapture(n); - ++NumAnnotated; + F.setDoesNotCapture(n); + ++NumAnnotated; + } + } + + void setOnlyReadsMemory(Function &F, unsigned n) { + if (!F.onlyReadsMemory(n)) { + F.setOnlyReadsMemory(n); + ++NumAnnotated; } } void setDoesNotAlias(Function &F, unsigned n) { if (!F.doesNotAlias(n)) { - F.setDoesNotAlias(n); - ++NumAnnotated; + F.setDoesNotAlias(n); + ++NumAnnotated; } } @@ -129,7 +136,8 @@ namespace { char FunctionAttrs::ID = 0; INITIALIZE_PASS_BEGIN(FunctionAttrs, "functionattrs", "Deduce function attributes", false, false) -INITIALIZE_AG_DEPENDENCY(CallGraph) +INITIALIZE_AG_DEPENDENCY(AliasAnalysis) +INITIALIZE_PASS_DEPENDENCY(CallGraph) INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfo) INITIALIZE_PASS_END(FunctionAttrs, "functionattrs", "Deduce function attributes", false, false) @@ -343,6 +351,7 @@ namespace { Function *F = CS.getCalledFunction(); if (!F || !SCCNodes.count(F)) { Captured = true; return true; } + bool Found = false; Function::arg_iterator AI = F->arg_begin(), AE = F->arg_end(); for (CallSite::arg_iterator PI = CS.arg_begin(), PE = CS.arg_end(); PI != PE; ++PI, ++AI) { @@ -353,10 +362,12 @@ namespace { } if (PI == U) { Uses.push_back(AI); + Found = true; break; } } - assert(!Uses.empty() && "Capturing call-site captured nothing?"); + assert(Found && "Capturing call-site captured nothing?"); + (void)Found; return false; } @@ -394,8 +405,100 @@ namespace llvm { }; } -/// AddNoCaptureAttrs - Deduce nocapture attributes for the SCC. -bool FunctionAttrs::AddNoCaptureAttrs(const CallGraphSCC &SCC) { +// Returns Attribute::None, Attribute::ReadOnly or Attribute::ReadNone. +static Attribute::AttrKind +determinePointerReadAttrs(Argument *A, + const SmallPtrSet<Argument*, 8> &SCCNodes) { + + SmallVector<Use*, 32> Worklist; + SmallSet<Use*, 32> Visited; + int Count = 0; + + bool IsRead = false; + // We don't need to track IsWritten. If A is written to, return immediately. + + for (Value::use_iterator UI = A->use_begin(), UE = A->use_end(); + UI != UE; ++UI) { + if (Count++ >= 20) + return Attribute::None; + + Use *U = &UI.getUse(); + Visited.insert(U); + Worklist.push_back(U); + } + + while (!Worklist.empty()) { + Use *U = Worklist.pop_back_val(); + Instruction *I = cast<Instruction>(U->getUser()); + Value *V = U->get(); + + switch (I->getOpcode()) { + case Instruction::BitCast: + case Instruction::GetElementPtr: + case Instruction::PHI: + case Instruction::Select: + // The original value is not read/written via this if the new value isn't. + for (Instruction::use_iterator UI = I->use_begin(), UE = I->use_end(); + UI != UE; ++UI) { + Use *U = &UI.getUse(); + if (Visited.insert(U)) + Worklist.push_back(U); + } + break; + + case Instruction::Call: + case Instruction::Invoke: { + CallSite CS(I); + if (CS.doesNotAccessMemory()) + continue; + + Function *F = CS.getCalledFunction(); + if (!F) { + if (CS.onlyReadsMemory()) { + IsRead = true; + continue; + } + return Attribute::None; + } + + Function::arg_iterator AI = F->arg_begin(), AE = F->arg_end(); + CallSite::arg_iterator B = CS.arg_begin(), E = CS.arg_end(); + for (CallSite::arg_iterator A = B; A != E; ++A, ++AI) { + if (A->get() == V) { + if (AI == AE) { + assert(F->isVarArg() && + "More params than args in non-varargs call."); + return Attribute::None; + } + if (SCCNodes.count(AI)) + continue; + if (!CS.onlyReadsMemory() && !CS.onlyReadsMemory(A - B)) + return Attribute::None; + if (!CS.doesNotAccessMemory(A - B)) + IsRead = true; + } + } + break; + } + + case Instruction::Load: + IsRead = true; + break; + + case Instruction::ICmp: + case Instruction::Ret: + break; + + default: + return Attribute::None; + } + } + + return IsRead ? Attribute::ReadOnly : Attribute::ReadNone; +} + +/// AddArgumentAttrs - Deduce nocapture attributes for the SCC. +bool FunctionAttrs::AddArgumentAttrs(const CallGraphSCC &SCC) { bool Changed = false; SmallPtrSet<Function*, 8> SCCNodes; @@ -442,8 +545,11 @@ bool FunctionAttrs::AddNoCaptureAttrs(const CallGraphSCC &SCC) { continue; } - for (Function::arg_iterator A = F->arg_begin(), E = F->arg_end(); A!=E; ++A) - if (A->getType()->isPointerTy() && !A->hasNoCaptureAttr()) { + for (Function::arg_iterator A = F->arg_begin(), E = F->arg_end(); + A != E; ++A) { + if (!A->getType()->isPointerTy()) continue; + bool HasNonLocalUses = false; + if (!A->hasNoCaptureAttr()) { ArgumentUsesTracker Tracker(SCCNodes); PointerMayBeCaptured(A, &Tracker); if (!Tracker.Captured) { @@ -458,12 +564,32 @@ bool FunctionAttrs::AddNoCaptureAttrs(const CallGraphSCC &SCC) { // its particulars for Argument-SCC analysis later. ArgumentGraphNode *Node = AG[A]; for (SmallVectorImpl<Argument*>::iterator UI = Tracker.Uses.begin(), - UE = Tracker.Uses.end(); UI != UE; ++UI) + UE = Tracker.Uses.end(); UI != UE; ++UI) { Node->Uses.push_back(AG[*UI]); + if (*UI != A) + HasNonLocalUses = true; + } } } // Otherwise, it's captured. Don't bother doing SCC analysis on it. } + if (!HasNonLocalUses && !A->onlyReadsMemory()) { + // Can we determine that it's readonly/readnone without doing an SCC? + // Note that we don't allow any calls at all here, or else our result + // will be dependent on the iteration order through the functions in the + // SCC. + SmallPtrSet<Argument*, 8> Self; + Self.insert(A); + Attribute::AttrKind R = determinePointerReadAttrs(A, Self); + if (R != Attribute::None) { + AttrBuilder B; + B.addAttribute(R); + A->addAttr(AttributeSet::get(A->getContext(), A->getArgNo() + 1, B)); + Changed = true; + R == Attribute::ReadOnly ? ++NumReadOnlyArg : ++NumReadNoneArg; + } + } + } } // The graph we've collected is partial because we stopped scanning for @@ -482,11 +608,8 @@ bool FunctionAttrs::AddNoCaptureAttrs(const CallGraphSCC &SCC) { // eg. "void f(int* x) { if (...) f(x); }" if (ArgumentSCC[0]->Uses.size() == 1 && ArgumentSCC[0]->Uses[0] == ArgumentSCC[0]) { - ArgumentSCC[0]-> - Definition-> - addAttr(AttributeSet::get(ArgumentSCC[0]->Definition->getContext(), - ArgumentSCC[0]->Definition->getArgNo() + 1, - B)); + Argument *A = ArgumentSCC[0]->Definition; + A->addAttr(AttributeSet::get(A->getContext(), A->getArgNo() + 1, B)); ++NumNoCapture; Changed = true; } @@ -532,6 +655,42 @@ bool FunctionAttrs::AddNoCaptureAttrs(const CallGraphSCC &SCC) { ++NumNoCapture; Changed = true; } + + // We also want to compute readonly/readnone. With a small number of false + // negatives, we can assume that any pointer which is captured isn't going + // to be provably readonly or readnone, since by definition we can't + // analyze all uses of a captured pointer. + // + // The false negatives happen when the pointer is captured by a function + // that promises readonly/readnone behaviour on the pointer, then the + // pointer's lifetime ends before anything that writes to arbitrary memory. + // Also, a readonly/readnone pointer may be returned, but returning a + // pointer is capturing it. + + Attribute::AttrKind ReadAttr = Attribute::ReadNone; + for (unsigned i = 0, e = ArgumentSCC.size(); i != e; ++i) { + Argument *A = ArgumentSCC[i]->Definition; + Attribute::AttrKind K = determinePointerReadAttrs(A, ArgumentSCCNodes); + if (K == Attribute::ReadNone) + continue; + if (K == Attribute::ReadOnly) { + ReadAttr = Attribute::ReadOnly; + continue; + } + ReadAttr = K; + break; + } + + if (ReadAttr != Attribute::None) { + AttrBuilder B; + B.addAttribute(ReadAttr); + for (unsigned i = 0, e = ArgumentSCC.size(); i != e; ++i) { + Argument *A = ArgumentSCC[i]->Definition; + A->addAttr(AttributeSet::get(A->getContext(), A->getArgNo() + 1, B)); + ReadAttr == Attribute::ReadOnly ? ++NumReadOnlyArg : ++NumReadNoneArg; + Changed = true; + } + } } return Changed; @@ -678,24 +837,32 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { setOnlyReadsMemory(F); setDoesNotThrow(F); break; - case LibFunc::strcpy: - case LibFunc::stpcpy: - case LibFunc::strcat: case LibFunc::strtol: case LibFunc::strtod: case LibFunc::strtof: case LibFunc::strtoul: case LibFunc::strtoll: case LibFunc::strtold: + case LibFunc::strtoull: + if (FTy->getNumParams() < 2 || + !FTy->getParamType(1)->isPointerTy()) + return false; + setDoesNotThrow(F); + setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 1); + break; + case LibFunc::strcpy: + case LibFunc::stpcpy: + case LibFunc::strcat: case LibFunc::strncat: case LibFunc::strncpy: case LibFunc::stpncpy: - case LibFunc::strtoull: if (FTy->getNumParams() < 2 || !FTy->getParamType(1)->isPointerTy()) return false; setDoesNotThrow(F); setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 2); break; case LibFunc::strxfrm: if (FTy->getNumParams() != 3 || @@ -705,14 +872,15 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { setDoesNotThrow(F); setDoesNotCapture(F, 1); setDoesNotCapture(F, 2); - break; - case LibFunc::strcmp: - case LibFunc::strspn: - case LibFunc::strncmp: - case LibFunc::strcspn: - case LibFunc::strcoll: - case LibFunc::strcasecmp: - case LibFunc::strncasecmp: + setOnlyReadsMemory(F, 2); + break; + case LibFunc::strcmp: //0,1 + case LibFunc::strspn: // 0,1 + case LibFunc::strncmp: // 0,1 + case LibFunc::strcspn: //0,1 + case LibFunc::strcoll: //0,1 + case LibFunc::strcasecmp: // 0,1 + case LibFunc::strncasecmp: // if (FTy->getNumParams() < 2 || !FTy->getParamType(0)->isPointerTy() || !FTy->getParamType(1)->isPointerTy()) @@ -736,8 +904,15 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { return false; setDoesNotThrow(F); setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 2); break; case LibFunc::scanf: + if (FTy->getNumParams() < 1 || !FTy->getParamType(0)->isPointerTy()) + return false; + setDoesNotThrow(F); + setDoesNotCapture(F, 1); + setOnlyReadsMemory(F, 1); + break; case LibFunc::setbuf: case LibFunc::setvbuf: if (FTy->getNumParams() < 1 || !FTy->getParamType(0)->isPointerTy()) @@ -753,11 +928,31 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { setDoesNotThrow(F); setDoesNotAlias(F, 0); setDoesNotCapture(F, 1); + setOnlyReadsMemory(F, 1); break; case LibFunc::stat: + case LibFunc::statvfs: + if (FTy->getNumParams() < 2 || + !FTy->getParamType(0)->isPointerTy() || + !FTy->getParamType(1)->isPointerTy()) + return false; + setDoesNotThrow(F); + setDoesNotCapture(F, 1); + setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 1); + break; case LibFunc::sscanf: + if (FTy->getNumParams() < 2 || + !FTy->getParamType(0)->isPointerTy() || + !FTy->getParamType(1)->isPointerTy()) + return false; + setDoesNotThrow(F); + setDoesNotCapture(F, 1); + setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 1); + setOnlyReadsMemory(F, 2); + break; case LibFunc::sprintf: - case LibFunc::statvfs: if (FTy->getNumParams() < 2 || !FTy->getParamType(0)->isPointerTy() || !FTy->getParamType(1)->isPointerTy()) @@ -765,6 +960,7 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { setDoesNotThrow(F); setDoesNotCapture(F, 1); setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 2); break; case LibFunc::snprintf: if (FTy->getNumParams() != 3 || @@ -774,6 +970,7 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { setDoesNotThrow(F); setDoesNotCapture(F, 1); setDoesNotCapture(F, 3); + setOnlyReadsMemory(F, 3); break; case LibFunc::setitimer: if (FTy->getNumParams() != 3 || @@ -783,6 +980,7 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { setDoesNotThrow(F); setDoesNotCapture(F, 2); setDoesNotCapture(F, 3); + setOnlyReadsMemory(F, 2); break; case LibFunc::system: if (FTy->getNumParams() != 1 || @@ -790,6 +988,7 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { return false; // May throw; "system" is a valid pthread cancellation point. setDoesNotCapture(F, 1); + setOnlyReadsMemory(F, 1); break; case LibFunc::malloc: if (FTy->getNumParams() != 1 || @@ -818,6 +1017,12 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { case LibFunc::modf: case LibFunc::modff: case LibFunc::modfl: + if (FTy->getNumParams() < 2 || + !FTy->getParamType(1)->isPointerTy()) + return false; + setDoesNotThrow(F); + setDoesNotCapture(F, 2); + break; case LibFunc::memcpy: case LibFunc::memccpy: case LibFunc::memmove: @@ -826,6 +1031,7 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { return false; setDoesNotThrow(F); setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 2); break; case LibFunc::memalign: if (!FTy->getReturnType()->isPointerTy()) @@ -833,6 +1039,13 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { setDoesNotAlias(F, 0); break; case LibFunc::mkdir: + if (FTy->getNumParams() == 0 || + !FTy->getParamType(0)->isPointerTy()) + return false; + setDoesNotThrow(F); + setDoesNotCapture(F, 1); + setOnlyReadsMemory(F, 1); + break; case LibFunc::mktime: if (FTy->getNumParams() == 0 || !FTy->getParamType(0)->isPointerTy()) @@ -856,8 +1069,14 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { // May throw; "read" is a valid pthread cancellation point. setDoesNotCapture(F, 2); break; - case LibFunc::rmdir: case LibFunc::rewind: + if (FTy->getNumParams() < 1 || + !FTy->getParamType(0)->isPointerTy()) + return false; + setDoesNotThrow(F); + setDoesNotCapture(F, 1); + break; + case LibFunc::rmdir: case LibFunc::remove: case LibFunc::realpath: if (FTy->getNumParams() < 1 || @@ -865,8 +1084,19 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { return false; setDoesNotThrow(F); setDoesNotCapture(F, 1); + setOnlyReadsMemory(F, 1); break; case LibFunc::rename: + if (FTy->getNumParams() < 2 || + !FTy->getParamType(0)->isPointerTy() || + !FTy->getParamType(1)->isPointerTy()) + return false; + setDoesNotThrow(F); + setDoesNotCapture(F, 1); + setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 1); + setOnlyReadsMemory(F, 2); + break; case LibFunc::readlink: if (FTy->getNumParams() < 2 || !FTy->getParamType(0)->isPointerTy() || @@ -875,12 +1105,14 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { setDoesNotThrow(F); setDoesNotCapture(F, 1); setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 1); break; case LibFunc::write: if (FTy->getNumParams() != 3 || !FTy->getParamType(1)->isPointerTy()) return false; // May throw; "write" is a valid pthread cancellation point. setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 2); break; case LibFunc::bcopy: if (FTy->getNumParams() != 3 || @@ -890,6 +1122,7 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { setDoesNotThrow(F); setDoesNotCapture(F, 1); setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 1); break; case LibFunc::bcmp: if (FTy->getNumParams() != 3 || @@ -916,6 +1149,12 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { break; case LibFunc::chmod: case LibFunc::chown: + if (FTy->getNumParams() == 0 || !FTy->getParamType(0)->isPointerTy()) + return false; + setDoesNotThrow(F); + setDoesNotCapture(F, 1); + setOnlyReadsMemory(F, 1); + break; case LibFunc::ctermid: case LibFunc::clearerr: case LibFunc::closedir: @@ -939,6 +1178,7 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { return false; setDoesNotThrow(F); setDoesNotCapture(F, 1); + setOnlyReadsMemory(F, 1); break; case LibFunc::fopen: if (FTy->getNumParams() != 2 || @@ -950,6 +1190,8 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { setDoesNotAlias(F, 0); setDoesNotCapture(F, 1); setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 1); + setOnlyReadsMemory(F, 2); break; case LibFunc::fdopen: if (FTy->getNumParams() != 2 || @@ -959,6 +1201,7 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { setDoesNotThrow(F); setDoesNotAlias(F, 0); setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 2); break; case LibFunc::feof: case LibFunc::free: @@ -1004,7 +1247,16 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { return false; setDoesNotThrow(F); setDoesNotCapture(F, 3); + break; case LibFunc::fread: + if (FTy->getNumParams() != 4 || + !FTy->getParamType(0)->isPointerTy() || + !FTy->getParamType(3)->isPointerTy()) + return false; + setDoesNotThrow(F); + setDoesNotCapture(F, 1); + setDoesNotCapture(F, 4); + break; case LibFunc::fwrite: if (FTy->getNumParams() != 4 || !FTy->getParamType(0)->isPointerTy() || @@ -1013,9 +1265,28 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { setDoesNotThrow(F); setDoesNotCapture(F, 1); setDoesNotCapture(F, 4); + break; case LibFunc::fputs: + if (FTy->getNumParams() < 2 || + !FTy->getParamType(0)->isPointerTy() || + !FTy->getParamType(1)->isPointerTy()) + return false; + setDoesNotThrow(F); + setDoesNotCapture(F, 1); + setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 1); + break; case LibFunc::fscanf: case LibFunc::fprintf: + if (FTy->getNumParams() < 2 || + !FTy->getParamType(0)->isPointerTy() || + !FTy->getParamType(1)->isPointerTy()) + return false; + setDoesNotThrow(F); + setDoesNotCapture(F, 1); + setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 2); + break; case LibFunc::fgetpos: if (FTy->getNumParams() < 2 || !FTy->getParamType(0)->isPointerTy() || @@ -1055,6 +1326,7 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { return false; setDoesNotThrow(F); setDoesNotCapture(F, 1); + setOnlyReadsMemory(F, 1); break; case LibFunc::ungetc: if (FTy->getNumParams() != 2 || !FTy->getParamType(1)->isPointerTy()) @@ -1063,12 +1335,24 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { setDoesNotCapture(F, 2); break; case LibFunc::uname: + if (FTy->getNumParams() != 1 || !FTy->getParamType(0)->isPointerTy()) + return false; + setDoesNotThrow(F); + setDoesNotCapture(F, 1); + break; case LibFunc::unlink: + if (FTy->getNumParams() != 1 || !FTy->getParamType(0)->isPointerTy()) + return false; + setDoesNotThrow(F); + setDoesNotCapture(F, 1); + setOnlyReadsMemory(F, 1); + break; case LibFunc::unsetenv: if (FTy->getNumParams() != 1 || !FTy->getParamType(0)->isPointerTy()) return false; setDoesNotThrow(F); setDoesNotCapture(F, 1); + setOnlyReadsMemory(F, 1); break; case LibFunc::utime: case LibFunc::utimes: @@ -1079,6 +1363,8 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { setDoesNotThrow(F); setDoesNotCapture(F, 1); setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 1); + setOnlyReadsMemory(F, 2); break; case LibFunc::putc: if (FTy->getNumParams() != 2 || !FTy->getParamType(1)->isPointerTy()) @@ -1093,13 +1379,20 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { return false; setDoesNotThrow(F); setDoesNotCapture(F, 1); + setOnlyReadsMemory(F, 1); break; case LibFunc::pread: + if (FTy->getNumParams() != 4 || !FTy->getParamType(1)->isPointerTy()) + return false; + // May throw; "pread" is a valid pthread cancellation point. + setDoesNotCapture(F, 2); + break; case LibFunc::pwrite: if (FTy->getNumParams() != 4 || !FTy->getParamType(1)->isPointerTy()) return false; - // May throw; these are valid pthread cancellation points. + // May throw; "pwrite" is a valid pthread cancellation point. setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 2); break; case LibFunc::putchar: setDoesNotThrow(F); @@ -1114,6 +1407,8 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { setDoesNotAlias(F, 0); setDoesNotCapture(F, 1); setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 1); + setOnlyReadsMemory(F, 2); break; case LibFunc::pclose: if (FTy->getNumParams() != 1 || !FTy->getParamType(0)->isPointerTy()) @@ -1126,8 +1421,19 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { return false; setDoesNotThrow(F); setDoesNotCapture(F, 1); + setOnlyReadsMemory(F, 1); break; case LibFunc::vsscanf: + if (FTy->getNumParams() != 3 || + !FTy->getParamType(1)->isPointerTy() || + !FTy->getParamType(2)->isPointerTy()) + return false; + setDoesNotThrow(F); + setDoesNotCapture(F, 1); + setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 1); + setOnlyReadsMemory(F, 2); + break; case LibFunc::vfscanf: if (FTy->getNumParams() != 3 || !FTy->getParamType(1)->isPointerTy() || @@ -1136,6 +1442,7 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { setDoesNotThrow(F); setDoesNotCapture(F, 1); setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 2); break; case LibFunc::valloc: if (!FTy->getReturnType()->isPointerTy()) @@ -1148,6 +1455,7 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { return false; setDoesNotThrow(F); setDoesNotCapture(F, 1); + setOnlyReadsMemory(F, 1); break; case LibFunc::vfprintf: case LibFunc::vsprintf: @@ -1158,6 +1466,7 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { setDoesNotThrow(F); setDoesNotCapture(F, 1); setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 2); break; case LibFunc::vsnprintf: if (FTy->getNumParams() != 4 || @@ -1167,12 +1476,14 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { setDoesNotThrow(F); setDoesNotCapture(F, 1); setDoesNotCapture(F, 3); + setOnlyReadsMemory(F, 3); break; case LibFunc::open: if (FTy->getNumParams() < 2 || !FTy->getParamType(0)->isPointerTy()) return false; // May throw; "open" is a valid pthread cancellation point. setDoesNotCapture(F, 1); + setOnlyReadsMemory(F, 1); break; case LibFunc::opendir: if (FTy->getNumParams() != 1 || @@ -1182,6 +1493,7 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { setDoesNotThrow(F); setDoesNotAlias(F, 0); setDoesNotCapture(F, 1); + setOnlyReadsMemory(F, 1); break; case LibFunc::tmpfile: if (!FTy->getReturnType()->isPointerTy()) @@ -1210,12 +1522,14 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { setDoesNotThrow(F); setDoesNotCapture(F, 1); setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 1); break; case LibFunc::lchown: if (FTy->getNumParams() != 3 || !FTy->getParamType(0)->isPointerTy()) return false; setDoesNotThrow(F); setDoesNotCapture(F, 1); + setOnlyReadsMemory(F, 1); break; case LibFunc::qsort: if (FTy->getNumParams() != 4 || !FTy->getParamType(3)->isPointerTy()) @@ -1232,6 +1546,7 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { setDoesNotThrow(F); setDoesNotAlias(F, 0); setDoesNotCapture(F, 1); + setOnlyReadsMemory(F, 1); break; case LibFunc::dunder_strtok_r: if (FTy->getNumParams() != 3 || @@ -1239,6 +1554,7 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { return false; setDoesNotThrow(F); setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 2); break; case LibFunc::under_IO_getc: if (FTy->getNumParams() != 1 || !FTy->getParamType(0)->isPointerTy()) @@ -1258,10 +1574,20 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { return false; setDoesNotThrow(F); setDoesNotCapture(F, 1); + setOnlyReadsMemory(F, 1); break; case LibFunc::stat64: case LibFunc::lstat64: case LibFunc::statvfs64: + if (FTy->getNumParams() < 1 || + !FTy->getParamType(0)->isPointerTy() || + !FTy->getParamType(1)->isPointerTy()) + return false; + setDoesNotThrow(F); + setDoesNotCapture(F, 1); + setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 1); + break; case LibFunc::dunder_isoc99_sscanf: if (FTy->getNumParams() < 1 || !FTy->getParamType(0)->isPointerTy() || @@ -1270,6 +1596,8 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { setDoesNotThrow(F); setDoesNotCapture(F, 1); setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 1); + setOnlyReadsMemory(F, 2); break; case LibFunc::fopen64: if (FTy->getNumParams() != 2 || @@ -1281,6 +1609,8 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { setDoesNotAlias(F, 0); setDoesNotCapture(F, 1); setDoesNotCapture(F, 2); + setOnlyReadsMemory(F, 1); + setOnlyReadsMemory(F, 2); break; case LibFunc::fseeko64: case LibFunc::ftello64: @@ -1307,7 +1637,18 @@ bool FunctionAttrs::inferPrototypeAttributes(Function &F) { return false; // May throw; "open" is a valid pthread cancellation point. setDoesNotCapture(F, 1); + setOnlyReadsMemory(F, 1); break; + case LibFunc::gettimeofday: + if (FTy->getNumParams() != 2 || !FTy->getParamType(0)->isPointerTy() || + !FTy->getParamType(1)->isPointerTy()) + return false; + // Currently some platforms have the restrict keyword on the arguments to + // gettimeofday. To be conservative, do not add noalias to gettimeofday's + // arguments. + setDoesNotThrow(F); + setDoesNotCapture(F, 1); + setDoesNotCapture(F, 2); default: // Didn't mark any attributes. return false; @@ -1339,7 +1680,7 @@ bool FunctionAttrs::runOnSCC(CallGraphSCC &SCC) { bool Changed = annotateLibraryCalls(SCC); Changed |= AddReadAttrs(SCC); - Changed |= AddNoCaptureAttrs(SCC); + Changed |= AddArgumentAttrs(SCC); Changed |= AddNoAliasAttrs(SCC); return Changed; } diff --git a/contrib/llvm/lib/Transforms/IPO/GlobalDCE.cpp b/contrib/llvm/lib/Transforms/IPO/GlobalDCE.cpp index 201f320..901295d 100644 --- a/contrib/llvm/lib/Transforms/IPO/GlobalDCE.cpp +++ b/contrib/llvm/lib/Transforms/IPO/GlobalDCE.cpp @@ -179,6 +179,9 @@ void GlobalDCE::GlobalIsNeeded(GlobalValue *G) { // any globals used will be marked as needed. Function *F = cast<Function>(G); + if (F->hasPrefixData()) + MarkUsedGlobalsAsNeeded(F->getPrefixData()); + for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) for (User::op_iterator U = I->op_begin(), E = I->op_end(); U != E; ++U) diff --git a/contrib/llvm/lib/Transforms/IPO/GlobalOpt.cpp b/contrib/llvm/lib/Transforms/IPO/GlobalOpt.cpp index 0ef900e..2ea89a1 100644 --- a/contrib/llvm/lib/Transforms/IPO/GlobalOpt.cpp +++ b/contrib/llvm/lib/Transforms/IPO/GlobalOpt.cpp @@ -37,7 +37,10 @@ #include "llvm/Support/GetElementPtrTypeIterator.h" #include "llvm/Support/MathExtras.h" #include "llvm/Support/raw_ostream.h" +#include "llvm/Support/ValueHandle.h" #include "llvm/Target/TargetLibraryInfo.h" +#include "llvm/Transforms/Utils/GlobalStatus.h" +#include "llvm/Transforms/Utils/ModuleUtils.h" #include <algorithm> using namespace llvm; @@ -59,7 +62,6 @@ STATISTIC(NumAliasesRemoved, "Number of global aliases eliminated"); STATISTIC(NumCXXDtorsRemoved, "Number of global C++ destructors removed"); namespace { - struct GlobalStatus; struct GlobalOpt : public ModulePass { virtual void getAnalysisUsage(AnalysisUsage &AU) const { AU.addRequired<TargetLibraryInfo>(); @@ -79,7 +81,6 @@ namespace { bool OptimizeGlobalCtorsList(GlobalVariable *&GCL); bool ProcessGlobal(GlobalVariable *GV,Module::global_iterator &GVI); bool ProcessInternalGlobal(GlobalVariable *GV,Module::global_iterator &GVI, - const SmallPtrSet<const PHINode*, 16> &PHIUsers, const GlobalStatus &GS); bool OptimizeEmptyGlobalCXXDtors(Function *CXAAtExitFn); @@ -97,209 +98,6 @@ INITIALIZE_PASS_END(GlobalOpt, "globalopt", ModulePass *llvm::createGlobalOptimizerPass() { return new GlobalOpt(); } -namespace { - -/// GlobalStatus - As we analyze each global, keep track of some information -/// about it. If we find out that the address of the global is taken, none of -/// this info will be accurate. -struct GlobalStatus { - /// isCompared - True if the global's address is used in a comparison. - bool isCompared; - - /// isLoaded - True if the global is ever loaded. If the global isn't ever - /// loaded it can be deleted. - bool isLoaded; - - /// StoredType - Keep track of what stores to the global look like. - /// - enum StoredType { - /// NotStored - There is no store to this global. It can thus be marked - /// constant. - NotStored, - - /// isInitializerStored - This global is stored to, but the only thing - /// stored is the constant it was initialized with. This is only tracked - /// for scalar globals. - isInitializerStored, - - /// isStoredOnce - This global is stored to, but only its initializer and - /// one other value is ever stored to it. If this global isStoredOnce, we - /// track the value stored to it in StoredOnceValue below. This is only - /// tracked for scalar globals. - isStoredOnce, - - /// isStored - This global is stored to by multiple values or something else - /// that we cannot track. - isStored - } StoredType; - - /// StoredOnceValue - If only one value (besides the initializer constant) is - /// ever stored to this global, keep track of what value it is. - Value *StoredOnceValue; - - /// AccessingFunction/HasMultipleAccessingFunctions - These start out - /// null/false. When the first accessing function is noticed, it is recorded. - /// When a second different accessing function is noticed, - /// HasMultipleAccessingFunctions is set to true. - const Function *AccessingFunction; - bool HasMultipleAccessingFunctions; - - /// HasNonInstructionUser - Set to true if this global has a user that is not - /// an instruction (e.g. a constant expr or GV initializer). - bool HasNonInstructionUser; - - /// AtomicOrdering - Set to the strongest atomic ordering requirement. - AtomicOrdering Ordering; - - GlobalStatus() : isCompared(false), isLoaded(false), StoredType(NotStored), - StoredOnceValue(0), AccessingFunction(0), - HasMultipleAccessingFunctions(false), - HasNonInstructionUser(false), Ordering(NotAtomic) {} -}; - -} - -/// StrongerOrdering - Return the stronger of the two ordering. If the two -/// orderings are acquire and release, then return AcquireRelease. -/// -static AtomicOrdering StrongerOrdering(AtomicOrdering X, AtomicOrdering Y) { - if (X == Acquire && Y == Release) return AcquireRelease; - if (Y == Acquire && X == Release) return AcquireRelease; - return (AtomicOrdering)std::max(X, Y); -} - -/// SafeToDestroyConstant - It is safe to destroy a constant iff it is only used -/// by constants itself. Note that constants cannot be cyclic, so this test is -/// pretty easy to implement recursively. -/// -static bool SafeToDestroyConstant(const Constant *C) { - if (isa<GlobalValue>(C)) return false; - - for (Value::const_use_iterator UI = C->use_begin(), E = C->use_end(); UI != E; - ++UI) - if (const Constant *CU = dyn_cast<Constant>(*UI)) { - if (!SafeToDestroyConstant(CU)) return false; - } else - return false; - return true; -} - - -/// AnalyzeGlobal - Look at all uses of the global and fill in the GlobalStatus -/// structure. If the global has its address taken, return true to indicate we -/// can't do anything with it. -/// -static bool AnalyzeGlobal(const Value *V, GlobalStatus &GS, - SmallPtrSet<const PHINode*, 16> &PHIUsers) { - for (Value::const_use_iterator UI = V->use_begin(), E = V->use_end(); UI != E; - ++UI) { - const User *U = *UI; - if (const ConstantExpr *CE = dyn_cast<ConstantExpr>(U)) { - GS.HasNonInstructionUser = true; - - // If the result of the constantexpr isn't pointer type, then we won't - // know to expect it in various places. Just reject early. - if (!isa<PointerType>(CE->getType())) return true; - - if (AnalyzeGlobal(CE, GS, PHIUsers)) return true; - } else if (const Instruction *I = dyn_cast<Instruction>(U)) { - if (!GS.HasMultipleAccessingFunctions) { - const Function *F = I->getParent()->getParent(); - if (GS.AccessingFunction == 0) - GS.AccessingFunction = F; - else if (GS.AccessingFunction != F) - GS.HasMultipleAccessingFunctions = true; - } - if (const LoadInst *LI = dyn_cast<LoadInst>(I)) { - GS.isLoaded = true; - // Don't hack on volatile loads. - if (LI->isVolatile()) return true; - GS.Ordering = StrongerOrdering(GS.Ordering, LI->getOrdering()); - } else if (const StoreInst *SI = dyn_cast<StoreInst>(I)) { - // Don't allow a store OF the address, only stores TO the address. - if (SI->getOperand(0) == V) return true; - - // Don't hack on volatile stores. - if (SI->isVolatile()) return true; - - GS.Ordering = StrongerOrdering(GS.Ordering, SI->getOrdering()); - - // If this is a direct store to the global (i.e., the global is a scalar - // value, not an aggregate), keep more specific information about - // stores. - if (GS.StoredType != GlobalStatus::isStored) { - if (const GlobalVariable *GV = dyn_cast<GlobalVariable>( - SI->getOperand(1))) { - Value *StoredVal = SI->getOperand(0); - - if (Constant *C = dyn_cast<Constant>(StoredVal)) { - if (C->isThreadDependent()) { - // The stored value changes between threads; don't track it. - return true; - } - } - - if (StoredVal == GV->getInitializer()) { - if (GS.StoredType < GlobalStatus::isInitializerStored) - GS.StoredType = GlobalStatus::isInitializerStored; - } else if (isa<LoadInst>(StoredVal) && - cast<LoadInst>(StoredVal)->getOperand(0) == GV) { - if (GS.StoredType < GlobalStatus::isInitializerStored) - GS.StoredType = GlobalStatus::isInitializerStored; - } else if (GS.StoredType < GlobalStatus::isStoredOnce) { - GS.StoredType = GlobalStatus::isStoredOnce; - GS.StoredOnceValue = StoredVal; - } else if (GS.StoredType == GlobalStatus::isStoredOnce && - GS.StoredOnceValue == StoredVal) { - // noop. - } else { - GS.StoredType = GlobalStatus::isStored; - } - } else { - GS.StoredType = GlobalStatus::isStored; - } - } - } else if (isa<BitCastInst>(I)) { - if (AnalyzeGlobal(I, GS, PHIUsers)) return true; - } else if (isa<GetElementPtrInst>(I)) { - if (AnalyzeGlobal(I, GS, PHIUsers)) return true; - } else if (isa<SelectInst>(I)) { - if (AnalyzeGlobal(I, GS, PHIUsers)) return true; - } else if (const PHINode *PN = dyn_cast<PHINode>(I)) { - // PHI nodes we can check just like select or GEP instructions, but we - // have to be careful about infinite recursion. - if (PHIUsers.insert(PN)) // Not already visited. - if (AnalyzeGlobal(I, GS, PHIUsers)) return true; - } else if (isa<CmpInst>(I)) { - GS.isCompared = true; - } else if (const MemTransferInst *MTI = dyn_cast<MemTransferInst>(I)) { - if (MTI->isVolatile()) return true; - if (MTI->getArgOperand(0) == V) - GS.StoredType = GlobalStatus::isStored; - if (MTI->getArgOperand(1) == V) - GS.isLoaded = true; - } else if (const MemSetInst *MSI = dyn_cast<MemSetInst>(I)) { - assert(MSI->getArgOperand(0) == V && "Memset only takes one pointer!"); - if (MSI->isVolatile()) return true; - GS.StoredType = GlobalStatus::isStored; - } else { - return true; // Any other non-load instruction might take address! - } - } else if (const Constant *C = dyn_cast<Constant>(U)) { - GS.HasNonInstructionUser = true; - // We might have a dead and dangling constant hanging off of here. - if (!SafeToDestroyConstant(C)) - return true; - } else { - GS.HasNonInstructionUser = true; - // Otherwise must be some other user. - return true; - } - } - - return false; -} - /// isLeakCheckerRoot - Is this global variable possibly used by a leak checker /// as a root? If so, we might not really want to eliminate the stores to it. static bool isLeakCheckerRoot(GlobalVariable *GV) { @@ -433,7 +231,7 @@ static bool CleanupPointerRootUsers(GlobalVariable *GV, Changed = true; } } else if (Constant *C = dyn_cast<Constant>(U)) { - if (SafeToDestroyConstant(C)) { + if (isSafeToDestroyConstant(C)) { C->destroyConstant(); // This could have invalidated UI, start over from scratch. Dead.clear(); @@ -470,9 +268,17 @@ static bool CleanupPointerRootUsers(GlobalVariable *GV, static bool CleanupConstantGlobalUsers(Value *V, Constant *Init, DataLayout *TD, TargetLibraryInfo *TLI) { bool Changed = false; - SmallVector<User*, 8> WorkList(V->use_begin(), V->use_end()); + // Note that we need to use a weak value handle for the worklist items. When + // we delete a constant array, we may also be holding pointer to one of its + // elements (or an element of one of its elements if we're dealing with an + // array of arrays) in the worklist. + SmallVector<WeakVH, 8> WorkList(V->use_begin(), V->use_end()); while (!WorkList.empty()) { - User *U = WorkList.pop_back_val(); + Value *UV = WorkList.pop_back_val(); + if (!UV) + continue; + + User *U = cast<User>(UV); if (LoadInst *LI = dyn_cast<LoadInst>(U)) { if (Init) { @@ -533,7 +339,7 @@ static bool CleanupConstantGlobalUsers(Value *V, Constant *Init, } else if (Constant *C = dyn_cast<Constant>(U)) { // If we have a chain of dead constantexprs or other things dangling from // us, and if they are all dead, nuke them without remorse. - if (SafeToDestroyConstant(C)) { + if (isSafeToDestroyConstant(C)) { C->destroyConstant(); CleanupConstantGlobalUsers(V, Init, TD, TLI); return true; @@ -548,7 +354,7 @@ static bool CleanupConstantGlobalUsers(Value *V, Constant *Init, static bool isSafeSROAElementUse(Value *V) { // We might have a dead and dangling constant hanging off of here. if (Constant *C = dyn_cast<Constant>(V)) - return SafeToDestroyConstant(C); + return isSafeToDestroyConstant(C); Instruction *I = dyn_cast<Instruction>(V); if (!I) return false; @@ -1372,8 +1178,7 @@ static Value *GetHeapSROAValue(Value *V, unsigned FieldNo, } else if (PHINode *PN = dyn_cast<PHINode>(V)) { // PN's type is pointer to struct. Make a new PHI of pointer to struct // field. - StructType *ST = - cast<StructType>(cast<PointerType>(PN->getType())->getElementType()); + StructType *ST = cast<StructType>(PN->getType()->getPointerElementType()); PHINode *NewPN = PHINode::Create(PointerType::getUnqual(ST->getElementType(FieldNo)), @@ -1504,7 +1309,7 @@ static GlobalVariable *PerformHeapAllocSRoA(GlobalVariable *GV, CallInst *CI, unsigned TypeSize = TD->getTypeAllocSize(FieldTy); if (StructType *ST = dyn_cast<StructType>(FieldTy)) TypeSize = TD->getStructLayout(ST)->getSizeInBytes(); - Type *IntPtrTy = TD->getIntPtrType(CI->getContext()); + Type *IntPtrTy = TD->getIntPtrType(CI->getType()); Value *NMI = CallInst::CreateMalloc(CI, IntPtrTy, FieldTy, ConstantInt::get(IntPtrTy, TypeSize), NElems, 0, @@ -1734,7 +1539,7 @@ static bool TryToOptimizeStoreOfMallocToGlobal(GlobalVariable *GV, // If this is a fixed size array, transform the Malloc to be an alloc of // structs. malloc [100 x struct],1 -> malloc struct, 100 if (ArrayType *AT = dyn_cast<ArrayType>(getMallocAllocatedType(CI, TLI))) { - Type *IntPtrTy = TD->getIntPtrType(CI->getContext()); + Type *IntPtrTy = TD->getIntPtrType(CI->getType()); unsigned TypeSize = TD->getStructLayout(AllocSTy)->getSizeInBytes(); Value *AllocSize = ConstantInt::get(IntPtrTy, TypeSize); Value *NumElements = ConstantInt::get(IntPtrTy, AT->getNumElements()); @@ -1916,13 +1721,12 @@ bool GlobalOpt::ProcessGlobal(GlobalVariable *GV, if (!GV->hasLocalLinkage()) return false; - SmallPtrSet<const PHINode*, 16> PHIUsers; GlobalStatus GS; - if (AnalyzeGlobal(GV, GS, PHIUsers)) + if (GlobalStatus::analyzeGlobal(GV, GS)) return false; - if (!GS.isCompared && !GV->hasUnnamedAddr()) { + if (!GS.IsCompared && !GV->hasUnnamedAddr()) { GV->setUnnamedAddr(true); NumUnnamed++; } @@ -1930,19 +1734,17 @@ bool GlobalOpt::ProcessGlobal(GlobalVariable *GV, if (GV->isConstant() || !GV->hasInitializer()) return false; - return ProcessInternalGlobal(GV, GVI, PHIUsers, GS); + return ProcessInternalGlobal(GV, GVI, GS); } /// ProcessInternalGlobal - Analyze the specified global variable and optimize /// it if possible. If we make a change, return true. bool GlobalOpt::ProcessInternalGlobal(GlobalVariable *GV, Module::global_iterator &GVI, - const SmallPtrSet<const PHINode*, 16> &PHIUsers, const GlobalStatus &GS) { // If this is a first class global and has only one accessing function - // and this function is main (which we know is not recursive we can make - // this global a local variable) we replace the global with a local alloca - // in this function. + // and this function is main (which we know is not recursive), we replace + // the global with a local alloca in this function. // // NOTE: It doesn't make sense to promote non single-value types since we // are just replacing static memory to stack memory. @@ -1971,7 +1773,7 @@ bool GlobalOpt::ProcessInternalGlobal(GlobalVariable *GV, // If the global is never loaded (but may be stored to), it is dead. // Delete it now. - if (!GS.isLoaded) { + if (!GS.IsLoaded) { DEBUG(dbgs() << "GLOBAL NEVER LOADED: " << *GV); bool Changed; @@ -1992,7 +1794,7 @@ bool GlobalOpt::ProcessInternalGlobal(GlobalVariable *GV, } return Changed; - } else if (GS.StoredType <= GlobalStatus::isInitializerStored) { + } else if (GS.StoredType <= GlobalStatus::InitializerStored) { DEBUG(dbgs() << "MARKING CONSTANT: " << *GV << "\n"); GV->setConstant(true); @@ -2015,7 +1817,7 @@ bool GlobalOpt::ProcessInternalGlobal(GlobalVariable *GV, GVI = FirstNewGV; // Don't skip the newly produced globals! return true; } - } else if (GS.StoredType == GlobalStatus::isStoredOnce) { + } else if (GS.StoredType == GlobalStatus::StoredOnce) { // If the initial value for the global was an undef value, and if only // one other value was stored into it, we can just change the // initializer to be the stored value, then delete all stores to the @@ -2048,11 +1850,14 @@ bool GlobalOpt::ProcessInternalGlobal(GlobalVariable *GV, // Otherwise, if the global was not a boolean, we can shrink it to be a // boolean. - if (Constant *SOVConstant = dyn_cast<Constant>(GS.StoredOnceValue)) - if (TryToShrinkGlobalToBoolean(GV, SOVConstant)) { - ++NumShrunkToBool; - return true; + if (Constant *SOVConstant = dyn_cast<Constant>(GS.StoredOnceValue)) { + if (GS.Ordering == NotAtomic) { + if (TryToShrinkGlobalToBoolean(GV, SOVConstant)) { + ++NumShrunkToBool; + return true; + } } + } } return false; @@ -2210,8 +2015,7 @@ static GlobalVariable *InstallGlobalCtors(GlobalVariable *GCL, CSVals[1] = 0; StructType *StructTy = - cast <StructType>( - cast<ArrayType>(GCL->getType()->getElementType())->getElementType()); + cast<StructType>(GCL->getType()->getElementType()->getArrayElementType()); // Create the new init list. std::vector<Constant*> CAList; @@ -2784,7 +2588,7 @@ bool Evaluator::EvaluateBlock(BasicBlock::iterator CurInst, Value *Ptr = PtrArg->stripPointerCasts(); if (GlobalVariable *GV = dyn_cast<GlobalVariable>(Ptr)) { Type *ElemTy = cast<PointerType>(GV->getType())->getElementType(); - if (!Size->isAllOnesValue() && + if (TD && !Size->isAllOnesValue() && Size->getValue().getLimitedValue() >= TD->getTypeStoreSize(ElemTy)) { Invariants.insert(GV); @@ -3041,107 +2845,148 @@ bool GlobalOpt::OptimizeGlobalCtorsList(GlobalVariable *&GCL) { return true; } -static Value::use_iterator getFirst(Value *V, SmallPtrSet<Use*, 8> &Tried) { - for (Value::use_iterator I = V->use_begin(), E = V->use_end(); I != E; ++I) { - Use *U = &I.getUse(); - if (Tried.count(U)) - continue; - - User *Usr = *I; - GlobalVariable *GV = dyn_cast<GlobalVariable>(Usr); - if (!GV || !GV->hasName()) { - Tried.insert(U); - return I; - } - - StringRef Name = GV->getName(); - if (Name != "llvm.used" && Name != "llvm.compiler_used") { - Tried.insert(U); - return I; - } - } - return V->use_end(); +static int compareNames(Constant *const *A, Constant *const *B) { + return (*A)->getName().compare((*B)->getName()); } -static bool replaceAllNonLLVMUsedUsesWith(Constant *Old, Constant *New); - -static bool replaceUsesOfWithOnConstant(ConstantArray *CA, Value *From, - Value *ToV, Use *U) { - Constant *To = cast<Constant>(ToV); - - SmallVector<Constant*, 8> NewOps; - for (unsigned i = 0, e = CA->getNumOperands(); i != e; ++i) { - Constant *Op = CA->getOperand(i); - NewOps.push_back(Op == From ? To : Op); +static void setUsedInitializer(GlobalVariable &V, + SmallPtrSet<GlobalValue *, 8> Init) { + if (Init.empty()) { + V.eraseFromParent(); + return; } - Constant *Replacement = ConstantArray::get(CA->getType(), NewOps); - assert(Replacement != CA && "CA didn't contain From!"); + SmallVector<llvm::Constant *, 8> UsedArray; + PointerType *Int8PtrTy = Type::getInt8PtrTy(V.getContext()); - bool Ret = replaceAllNonLLVMUsedUsesWith(CA, Replacement); - if (Replacement->use_empty()) - Replacement->destroyConstant(); - if (CA->use_empty()) - CA->destroyConstant(); - return Ret; + for (SmallPtrSet<GlobalValue *, 8>::iterator I = Init.begin(), E = Init.end(); + I != E; ++I) { + Constant *Cast = llvm::ConstantExpr::getBitCast(*I, Int8PtrTy); + UsedArray.push_back(Cast); + } + // Sort to get deterministic order. + array_pod_sort(UsedArray.begin(), UsedArray.end(), compareNames); + ArrayType *ATy = ArrayType::get(Int8PtrTy, UsedArray.size()); + + Module *M = V.getParent(); + V.removeFromParent(); + GlobalVariable *NV = + new GlobalVariable(*M, ATy, false, llvm::GlobalValue::AppendingLinkage, + llvm::ConstantArray::get(ATy, UsedArray), ""); + NV->takeName(&V); + NV->setSection("llvm.metadata"); + delete &V; } -static bool replaceUsesOfWithOnConstant(ConstantExpr *CE, Value *From, - Value *ToV, Use *U) { - Constant *To = cast<Constant>(ToV); - SmallVector<Constant*, 8> NewOps; - for (unsigned i = 0, e = CE->getNumOperands(); i != e; ++i) { - Constant *Op = CE->getOperand(i); - NewOps.push_back(Op == From ? To : Op); +namespace { +/// \brief An easy to access representation of llvm.used and llvm.compiler.used. +class LLVMUsed { + SmallPtrSet<GlobalValue *, 8> Used; + SmallPtrSet<GlobalValue *, 8> CompilerUsed; + GlobalVariable *UsedV; + GlobalVariable *CompilerUsedV; + +public: + LLVMUsed(Module &M) { + UsedV = collectUsedGlobalVariables(M, Used, false); + CompilerUsedV = collectUsedGlobalVariables(M, CompilerUsed, true); + } + typedef SmallPtrSet<GlobalValue *, 8>::iterator iterator; + iterator usedBegin() { return Used.begin(); } + iterator usedEnd() { return Used.end(); } + iterator compilerUsedBegin() { return CompilerUsed.begin(); } + iterator compilerUsedEnd() { return CompilerUsed.end(); } + bool usedCount(GlobalValue *GV) const { return Used.count(GV); } + bool compilerUsedCount(GlobalValue *GV) const { + return CompilerUsed.count(GV); + } + bool usedErase(GlobalValue *GV) { return Used.erase(GV); } + bool compilerUsedErase(GlobalValue *GV) { return CompilerUsed.erase(GV); } + bool usedInsert(GlobalValue *GV) { return Used.insert(GV); } + bool compilerUsedInsert(GlobalValue *GV) { return CompilerUsed.insert(GV); } + + void syncVariablesAndSets() { + if (UsedV) + setUsedInitializer(*UsedV, Used); + if (CompilerUsedV) + setUsedInitializer(*CompilerUsedV, CompilerUsed); } +}; +} - Constant *Replacement = CE->getWithOperands(NewOps); - assert(Replacement != CE && "CE didn't contain From!"); +static bool hasUseOtherThanLLVMUsed(GlobalAlias &GA, const LLVMUsed &U) { + if (GA.use_empty()) // No use at all. + return false; - bool Ret = replaceAllNonLLVMUsedUsesWith(CE, Replacement); - if (Replacement->use_empty()) - Replacement->destroyConstant(); - if (CE->use_empty()) - CE->destroyConstant(); - return Ret; + assert((!U.usedCount(&GA) || !U.compilerUsedCount(&GA)) && + "We should have removed the duplicated " + "element from llvm.compiler.used"); + if (!GA.hasOneUse()) + // Strictly more than one use. So at least one is not in llvm.used and + // llvm.compiler.used. + return true; + + // Exactly one use. Check if it is in llvm.used or llvm.compiler.used. + return !U.usedCount(&GA) && !U.compilerUsedCount(&GA); } -static bool replaceUsesOfWithOnConstant(Constant *C, Value *From, Value *To, - Use *U) { - if (ConstantArray *CA = dyn_cast<ConstantArray>(C)) - return replaceUsesOfWithOnConstant(CA, From, To, U); - if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C)) - return replaceUsesOfWithOnConstant(CE, From, To, U); - C->replaceUsesOfWithOnConstant(From, To, U); - return true; +static bool hasMoreThanOneUseOtherThanLLVMUsed(GlobalValue &V, + const LLVMUsed &U) { + unsigned N = 2; + assert((!U.usedCount(&V) || !U.compilerUsedCount(&V)) && + "We should have removed the duplicated " + "element from llvm.compiler.used"); + if (U.usedCount(&V) || U.compilerUsedCount(&V)) + ++N; + return V.hasNUsesOrMore(N); } -static bool replaceAllNonLLVMUsedUsesWith(Constant *Old, Constant *New) { - SmallPtrSet<Use*, 8> Tried; - bool Ret = false; - for (;;) { - Value::use_iterator I = getFirst(Old, Tried); - if (I == Old->use_end()) - break; - Use &U = I.getUse(); +static bool mayHaveOtherReferences(GlobalAlias &GA, const LLVMUsed &U) { + if (!GA.hasLocalLinkage()) + return true; - // Must handle Constants specially, we cannot call replaceUsesOfWith on a - // constant because they are uniqued. - if (Constant *C = dyn_cast<Constant>(U.getUser())) { - if (!isa<GlobalValue>(C)) { - Ret |= replaceUsesOfWithOnConstant(C, Old, New, &U); - continue; - } - } + return U.usedCount(&GA) || U.compilerUsedCount(&GA); +} - U.set(New); +static bool hasUsesToReplace(GlobalAlias &GA, LLVMUsed &U, bool &RenameTarget) { + RenameTarget = false; + bool Ret = false; + if (hasUseOtherThanLLVMUsed(GA, U)) Ret = true; - } - return Ret; + + // If the alias is externally visible, we may still be able to simplify it. + if (!mayHaveOtherReferences(GA, U)) + return Ret; + + // If the aliasee has internal linkage, give it the name and linkage + // of the alias, and delete the alias. This turns: + // define internal ... @f(...) + // @a = alias ... @f + // into: + // define ... @a(...) + Constant *Aliasee = GA.getAliasee(); + GlobalValue *Target = cast<GlobalValue>(Aliasee->stripPointerCasts()); + if (!Target->hasLocalLinkage()) + return Ret; + + // Do not perform the transform if multiple aliases potentially target the + // aliasee. This check also ensures that it is safe to replace the section + // and other attributes of the aliasee with those of the alias. + if (hasMoreThanOneUseOtherThanLLVMUsed(*Target, U)) + return Ret; + + RenameTarget = true; + return true; } bool GlobalOpt::OptimizeGlobalAliases(Module &M) { bool Changed = false; + LLVMUsed Used(M); + + for (SmallPtrSet<GlobalValue *, 8>::iterator I = Used.usedBegin(), + E = Used.usedEnd(); + I != E; ++I) + Used.compilerUsedErase(*I); for (Module::alias_iterator I = M.alias_begin(), E = M.alias_end(); I != E;) { @@ -3156,38 +3001,29 @@ bool GlobalOpt::OptimizeGlobalAliases(Module &M) { Constant *Aliasee = J->getAliasee(); GlobalValue *Target = cast<GlobalValue>(Aliasee->stripPointerCasts()); Target->removeDeadConstantUsers(); - bool hasOneUse = Target->hasOneUse() && Aliasee->hasOneUse(); // Make all users of the alias use the aliasee instead. - if (replaceAllNonLLVMUsedUsesWith(J, Aliasee)) { - ++NumAliasesResolved; - Changed = true; - } - if (!J->use_empty()) + bool RenameTarget; + if (!hasUsesToReplace(*J, Used, RenameTarget)) continue; - // If the alias is externally visible, we may still be able to simplify it. - if (!J->hasLocalLinkage()) { - // If the aliasee has internal linkage, give it the name and linkage - // of the alias, and delete the alias. This turns: - // define internal ... @f(...) - // @a = alias ... @f - // into: - // define ... @a(...) - if (!Target->hasLocalLinkage()) - continue; - - // Do not perform the transform if multiple aliases potentially target the - // aliasee. This check also ensures that it is safe to replace the section - // and other attributes of the aliasee with those of the alias. - if (!hasOneUse) - continue; + J->replaceAllUsesWith(Aliasee); + ++NumAliasesResolved; + Changed = true; + if (RenameTarget) { // Give the aliasee the name, linkage and other attributes of the alias. Target->takeName(J); Target->setLinkage(J->getLinkage()); Target->GlobalValue::copyAttributesFrom(J); - } + + if (Used.usedErase(J)) + Used.usedInsert(Target); + + if (Used.compilerUsedErase(J)) + Used.compilerUsedInsert(Target); + } else if (mayHaveOtherReferences(*J, Used)) + continue; // Delete the alias. M.getAliasList().erase(J); @@ -3195,6 +3031,8 @@ bool GlobalOpt::OptimizeGlobalAliases(Module &M) { Changed = true; } + Used.syncVariablesAndSets(); + return Changed; } @@ -3323,8 +3161,6 @@ bool GlobalOpt::runOnModule(Module &M) { // Try to find the llvm.globalctors list. GlobalVariable *GlobalCtors = FindGlobalCtors(M); - Function *CXAAtExitFn = FindCXAAtExit(M, TLI); - bool LocalChange = true; while (LocalChange) { LocalChange = false; @@ -3342,7 +3178,9 @@ bool GlobalOpt::runOnModule(Module &M) { // Resolve aliases, when possible. LocalChange |= OptimizeGlobalAliases(M); - // Try to remove trivial global destructors. + // Try to remove trivial global destructors if they are not removed + // already. + Function *CXAAtExitFn = FindCXAAtExit(M, TLI); if (CXAAtExitFn) LocalChange |= OptimizeEmptyGlobalCXXDtors(CXAAtExitFn); diff --git a/contrib/llvm/lib/Transforms/IPO/InlineAlways.cpp b/contrib/llvm/lib/Transforms/IPO/InlineAlways.cpp index a0095da..437597e 100644 --- a/contrib/llvm/lib/Transforms/IPO/InlineAlways.cpp +++ b/contrib/llvm/lib/Transforms/IPO/InlineAlways.cpp @@ -63,7 +63,7 @@ public: char AlwaysInliner::ID = 0; INITIALIZE_PASS_BEGIN(AlwaysInliner, "always-inline", "Inliner for always_inline functions", false, false) -INITIALIZE_AG_DEPENDENCY(CallGraph) +INITIALIZE_PASS_DEPENDENCY(CallGraph) INITIALIZE_PASS_DEPENDENCY(InlineCostAnalysis) INITIALIZE_PASS_END(AlwaysInliner, "always-inline", "Inliner for always_inline functions", false, false) diff --git a/contrib/llvm/lib/Transforms/IPO/InlineSimple.cpp b/contrib/llvm/lib/Transforms/IPO/InlineSimple.cpp index a4f7026..57379a3 100644 --- a/contrib/llvm/lib/Transforms/IPO/InlineSimple.cpp +++ b/contrib/llvm/lib/Transforms/IPO/InlineSimple.cpp @@ -28,7 +28,7 @@ using namespace llvm; namespace { -/// \brief Actaul inliner pass implementation. +/// \brief Actual inliner pass implementation. /// /// The common implementation of the inlining logic is shared between this /// inliner pass and the always inliner pass. The two passes use different cost @@ -61,7 +61,7 @@ public: char SimpleInliner::ID = 0; INITIALIZE_PASS_BEGIN(SimpleInliner, "inline", "Function Integration/Inlining", false, false) -INITIALIZE_AG_DEPENDENCY(CallGraph) +INITIALIZE_PASS_DEPENDENCY(CallGraph) INITIALIZE_PASS_DEPENDENCY(InlineCostAnalysis) INITIALIZE_PASS_END(SimpleInliner, "inline", "Function Integration/Inlining", false, false) diff --git a/contrib/llvm/lib/Transforms/IPO/Inliner.cpp b/contrib/llvm/lib/Transforms/IPO/Inliner.cpp index 663ddb7..d75d6ca 100644 --- a/contrib/llvm/lib/Transforms/IPO/Inliner.cpp +++ b/contrib/llvm/lib/Transforms/IPO/Inliner.cpp @@ -116,7 +116,8 @@ static void AdjustCallerSSPLevel(Function *Caller, Function *Callee) { /// any new allocas to the set if not possible. static bool InlineCallIfPossible(CallSite CS, InlineFunctionInfo &IFI, InlinedArrayAllocasTy &InlinedArrayAllocas, - int InlineHistory, bool InsertLifetime) { + int InlineHistory, bool InsertLifetime, + const DataLayout *TD) { Function *Callee = CS.getCalledFunction(); Function *Caller = CS.getCaller(); @@ -189,6 +190,14 @@ static bool InlineCallIfPossible(CallSite CS, InlineFunctionInfo &IFI, bool MergedAwayAlloca = false; for (unsigned i = 0, e = AllocasForType.size(); i != e; ++i) { AllocaInst *AvailableAlloca = AllocasForType[i]; + + unsigned Align1 = AI->getAlignment(), + Align2 = AvailableAlloca->getAlignment(); + // If we don't have data layout information, and only one alloca is using + // the target default, then we can't safely merge them because we can't + // pick the greater alignment. + if (!TD && (!Align1 || !Align2) && Align1 != Align2) + continue; // The available alloca has to be in the right function, not in some other // function in this SCC. @@ -206,6 +215,20 @@ static bool InlineCallIfPossible(CallSite CS, InlineFunctionInfo &IFI, << *AvailableAlloca << '\n'); AI->replaceAllUsesWith(AvailableAlloca); + + if (Align1 != Align2) { + if (!Align1 || !Align2) { + assert(TD && "DataLayout required to compare default alignments"); + unsigned TypeAlign = TD->getABITypeAlignment(AI->getAllocatedType()); + + Align1 = Align1 ? Align1 : TypeAlign; + Align2 = Align2 ? Align2 : TypeAlign; + } + + if (Align1 > Align2) + AvailableAlloca->setAlignment(AI->getAlignment()); + } + AI->eraseFromParent(); MergedAwayAlloca = true; ++NumMergedAllocas; @@ -482,7 +505,7 @@ bool Inliner::runOnSCC(CallGraphSCC &SCC) { // Attempt to inline the function. if (!InlineCallIfPossible(CS, InlineInfo, InlinedArrayAllocas, - InlineHistoryID, InsertLifetime)) + InlineHistoryID, InsertLifetime, TD)) continue; ++NumInlined; diff --git a/contrib/llvm/lib/Transforms/IPO/Internalize.cpp b/contrib/llvm/lib/Transforms/IPO/Internalize.cpp index 4bfab5b..64e2ced 100644 --- a/contrib/llvm/lib/Transforms/IPO/Internalize.cpp +++ b/contrib/llvm/lib/Transforms/IPO/Internalize.cpp @@ -11,10 +11,17 @@ // If the function or variable is not in the list of external names given to // the pass it is marked as internal. // +// This transformation would not be legal in a regular compilation, but it gets +// extra information from the linker about what is safe. +// +// For example: Internalizing a function with external linkage. Only if we are +// told it is only used from within this module, it is safe to do it. +// //===----------------------------------------------------------------------===// #define DEBUG_TYPE "internalize" #include "llvm/Transforms/IPO.h" +#include "llvm/ADT/SmallPtrSet.h" #include "llvm/ADT/Statistic.h" #include "llvm/Analysis/CallGraph.h" #include "llvm/IR/Module.h" @@ -22,6 +29,8 @@ #include "llvm/Support/CommandLine.h" #include "llvm/Support/Debug.h" #include "llvm/Support/raw_ostream.h" +#include "llvm/Transforms/Utils/GlobalStatus.h" +#include "llvm/Transforms/Utils/ModuleUtils.h" #include <fstream> #include <set> using namespace llvm; @@ -48,10 +57,8 @@ namespace { public: static char ID; // Pass identification, replacement for typeid explicit InternalizePass(); - explicit InternalizePass(ArrayRef<const char *> exportList); + explicit InternalizePass(ArrayRef<const char *> ExportList); void LoadFile(const char *Filename); - void ClearExportList(); - void AddToExportList(const std::string &val); virtual bool runOnModule(Module &M); virtual void getAnalysisUsage(AnalysisUsage &AU) const { @@ -70,15 +77,14 @@ InternalizePass::InternalizePass() initializeInternalizePassPass(*PassRegistry::getPassRegistry()); if (!APIFile.empty()) // If a filename is specified, use it. LoadFile(APIFile.c_str()); - if (!APIList.empty()) // If a list is specified, use it as well. - ExternalNames.insert(APIList.begin(), APIList.end()); + ExternalNames.insert(APIList.begin(), APIList.end()); } -InternalizePass::InternalizePass(ArrayRef<const char *> exportList) +InternalizePass::InternalizePass(ArrayRef<const char *> ExportList) : ModulePass(ID){ initializeInternalizePassPass(*PassRegistry::getPassRegistry()); - for(ArrayRef<const char *>::const_iterator itr = exportList.begin(); - itr != exportList.end(); itr++) { + for(ArrayRef<const char *>::const_iterator itr = ExportList.begin(); + itr != ExportList.end(); itr++) { ExternalNames.insert(*itr); } } @@ -99,12 +105,25 @@ void InternalizePass::LoadFile(const char *Filename) { } } -void InternalizePass::ClearExportList() { - ExternalNames.clear(); -} +static bool shouldInternalize(const GlobalValue &GV, + const std::set<std::string> &ExternalNames) { + // Function must be defined here + if (GV.isDeclaration()) + return false; + + // Available externally is really just a "declaration with a body". + if (GV.hasAvailableExternallyLinkage()) + return false; + + // Already has internal linkage + if (GV.hasLocalLinkage()) + return false; + + // Marked to keep external? + if (ExternalNames.count(GV.getName())) + return false; -void InternalizePass::AddToExportList(const std::string &val) { - ExternalNames.insert(val); + return true; } bool InternalizePass::runOnModule(Module &M) { @@ -112,26 +131,40 @@ bool InternalizePass::runOnModule(Module &M) { CallGraphNode *ExternalNode = CG ? CG->getExternalCallingNode() : 0; bool Changed = false; - // Never internalize functions which code-gen might insert. - // FIXME: We should probably add this (and the __stack_chk_guard) via some - // type of call-back in CodeGen. - ExternalNames.insert("__stack_chk_fail"); + SmallPtrSet<GlobalValue *, 8> Used; + collectUsedGlobalVariables(M, Used, false); + + // We must assume that globals in llvm.used have a reference that not even + // the linker can see, so we don't internalize them. + // For llvm.compiler.used the situation is a bit fuzzy. The assembler and + // linker can drop those symbols. If this pass is running as part of LTO, + // one might think that it could just drop llvm.compiler.used. The problem + // is that even in LTO llvm doesn't see every reference. For example, + // we don't see references from function local inline assembly. To be + // conservative, we internalize symbols in llvm.compiler.used, but we + // keep llvm.compiler.used so that the symbol is not deleted by llvm. + for (SmallPtrSet<GlobalValue *, 8>::iterator I = Used.begin(), E = Used.end(); + I != E; ++I) { + GlobalValue *V = *I; + ExternalNames.insert(V->getName()); + } // Mark all functions not in the api as internal. // FIXME: maybe use private linkage? - for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I) - if (!I->isDeclaration() && // Function must be defined here - // Available externally is really just a "declaration with a body". - !I->hasAvailableExternallyLinkage() && - !I->hasLocalLinkage() && // Can't already have internal linkage - !ExternalNames.count(I->getName())) {// Not marked to keep external? - I->setLinkage(GlobalValue::InternalLinkage); + for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I) { + if (!shouldInternalize(*I, ExternalNames)) + continue; + + I->setLinkage(GlobalValue::InternalLinkage); + + if (ExternalNode) // Remove a callgraph edge from the external node to this function. - if (ExternalNode) ExternalNode->removeOneAbstractEdgeTo((*CG)[I]); - Changed = true; - ++NumFunctions; - DEBUG(dbgs() << "Internalizing func " << I->getName() << "\n"); - } + ExternalNode->removeOneAbstractEdgeTo((*CG)[I]); + + Changed = true; + ++NumFunctions; + DEBUG(dbgs() << "Internalizing func " << I->getName() << "\n"); + } // Never internalize the llvm.used symbol. It is used to implement // attribute((used)). @@ -146,35 +179,36 @@ bool InternalizePass::runOnModule(Module &M) { ExternalNames.insert("llvm.global.annotations"); // Never internalize symbols code-gen inserts. + // FIXME: We should probably add this (and the __stack_chk_guard) via some + // type of call-back in CodeGen. + ExternalNames.insert("__stack_chk_fail"); ExternalNames.insert("__stack_chk_guard"); // Mark all global variables with initializers that are not in the api as // internal as well. // FIXME: maybe use private linkage? for (Module::global_iterator I = M.global_begin(), E = M.global_end(); - I != E; ++I) - if (!I->isDeclaration() && !I->hasLocalLinkage() && - // Available externally is really just a "declaration with a body". - !I->hasAvailableExternallyLinkage() && - !ExternalNames.count(I->getName())) { - I->setLinkage(GlobalValue::InternalLinkage); - Changed = true; - ++NumGlobals; - DEBUG(dbgs() << "Internalized gvar " << I->getName() << "\n"); - } + I != E; ++I) { + if (!shouldInternalize(*I, ExternalNames)) + continue; + + I->setLinkage(GlobalValue::InternalLinkage); + Changed = true; + ++NumGlobals; + DEBUG(dbgs() << "Internalized gvar " << I->getName() << "\n"); + } // Mark all aliases that are not in the api as internal as well. for (Module::alias_iterator I = M.alias_begin(), E = M.alias_end(); - I != E; ++I) - if (!I->isDeclaration() && !I->hasInternalLinkage() && - // Available externally is really just a "declaration with a body". - !I->hasAvailableExternallyLinkage() && - !ExternalNames.count(I->getName())) { - I->setLinkage(GlobalValue::InternalLinkage); - Changed = true; - ++NumAliases; - DEBUG(dbgs() << "Internalized alias " << I->getName() << "\n"); - } + I != E; ++I) { + if (!shouldInternalize(*I, ExternalNames)) + continue; + + I->setLinkage(GlobalValue::InternalLinkage); + Changed = true; + ++NumAliases; + DEBUG(dbgs() << "Internalized alias " << I->getName() << "\n"); + } return Changed; } @@ -183,6 +217,6 @@ ModulePass *llvm::createInternalizePass() { return new InternalizePass(); } -ModulePass *llvm::createInternalizePass(ArrayRef<const char *> el) { - return new InternalizePass(el); +ModulePass *llvm::createInternalizePass(ArrayRef<const char *> ExportList) { + return new InternalizePass(ExportList); } diff --git a/contrib/llvm/lib/Transforms/IPO/MergeFunctions.cpp b/contrib/llvm/lib/Transforms/IPO/MergeFunctions.cpp index 4ce749c..3861421 100644 --- a/contrib/llvm/lib/Transforms/IPO/MergeFunctions.cpp +++ b/contrib/llvm/lib/Transforms/IPO/MergeFunctions.cpp @@ -210,16 +210,20 @@ private: // Any two pointers in the same address space are equivalent, intptr_t and // pointers are equivalent. Otherwise, standard type equivalence rules apply. bool FunctionComparator::isEquivalentType(Type *Ty1, Type *Ty2) const { + + PointerType *PTy1 = dyn_cast<PointerType>(Ty1); + PointerType *PTy2 = dyn_cast<PointerType>(Ty2); + + if (TD) { + if (PTy1 && PTy1->getAddressSpace() == 0) Ty1 = TD->getIntPtrType(Ty1); + if (PTy2 && PTy2->getAddressSpace() == 0) Ty2 = TD->getIntPtrType(Ty2); + } + if (Ty1 == Ty2) return true; - if (Ty1->getTypeID() != Ty2->getTypeID()) { - if (TD) { - LLVMContext &Ctx = Ty1->getContext(); - if (isa<PointerType>(Ty1) && Ty2 == TD->getIntPtrType(Ctx)) return true; - if (isa<PointerType>(Ty2) && Ty1 == TD->getIntPtrType(Ctx)) return true; - } + + if (Ty1->getTypeID() != Ty2->getTypeID()) return false; - } switch (Ty1->getTypeID()) { default: @@ -241,8 +245,7 @@ bool FunctionComparator::isEquivalentType(Type *Ty1, Type *Ty2) const { return true; case Type::PointerTyID: { - PointerType *PTy1 = cast<PointerType>(Ty1); - PointerType *PTy2 = cast<PointerType>(Ty2); + assert(PTy1 && PTy2 && "Both types must be pointers here."); return PTy1->getAddressSpace() == PTy2->getAddressSpace(); } @@ -352,14 +355,19 @@ bool FunctionComparator::isEquivalentOperation(const Instruction *I1, // Determine whether two GEP operations perform the same underlying arithmetic. bool FunctionComparator::isEquivalentGEP(const GEPOperator *GEP1, const GEPOperator *GEP2) { - // When we have target data, we can reduce the GEP down to the value in bytes - // added to the address. - unsigned BitWidth = TD ? TD->getPointerSizeInBits() : 1; - APInt Offset1(BitWidth, 0), Offset2(BitWidth, 0); - if (TD && - GEP1->accumulateConstantOffset(*TD, Offset1) && - GEP2->accumulateConstantOffset(*TD, Offset2)) { - return Offset1 == Offset2; + unsigned AS = GEP1->getPointerAddressSpace(); + if (AS != GEP2->getPointerAddressSpace()) + return false; + + if (TD) { + // When we have target data, we can reduce the GEP down to the value in bytes + // added to the address. + unsigned BitWidth = TD ? TD->getPointerSizeInBits(AS) : 1; + APInt Offset1(BitWidth, 0), Offset2(BitWidth, 0); + if (GEP1->accumulateConstantOffset(*TD, Offset1) && + GEP2->accumulateConstantOffset(*TD, Offset2)) { + return Offset1 == Offset2; + } } if (GEP1->getPointerOperand()->getType() != @@ -713,6 +721,19 @@ void MergeFunctions::writeThunkOrAlias(Function *F, Function *G) { writeThunk(F, G); } +// Helper for writeThunk, +// Selects proper bitcast operation, +// but a bit simplier then CastInst::getCastOpcode. +static Value* createCast(IRBuilder<false> &Builder, Value *V, Type *DestTy) { + Type *SrcTy = V->getType(); + if (SrcTy->isIntegerTy() && DestTy->isPointerTy()) + return Builder.CreateIntToPtr(V, DestTy); + else if (SrcTy->isPointerTy() && DestTy->isIntegerTy()) + return Builder.CreatePtrToInt(V, DestTy); + else + return Builder.CreateBitCast(V, DestTy); +} + // Replace G with a simple tail call to bitcast(F). Also replace direct uses // of G with bitcast(F). Deletes G. void MergeFunctions::writeThunk(Function *F, Function *G) { @@ -738,7 +759,7 @@ void MergeFunctions::writeThunk(Function *F, Function *G) { FunctionType *FFTy = F->getFunctionType(); for (Function::arg_iterator AI = NewG->arg_begin(), AE = NewG->arg_end(); AI != AE; ++AI) { - Args.push_back(Builder.CreateBitCast(AI, FFTy->getParamType(i))); + Args.push_back(createCast(Builder, (Value*)AI, FFTy->getParamType(i))); ++i; } @@ -748,13 +769,7 @@ void MergeFunctions::writeThunk(Function *F, Function *G) { if (NewG->getReturnType()->isVoidTy()) { Builder.CreateRetVoid(); } else { - Type *RetTy = NewG->getReturnType(); - if (CI->getType()->isIntegerTy() && RetTy->isPointerTy()) - Builder.CreateRet(Builder.CreateIntToPtr(CI, RetTy)); - else if (CI->getType()->isPointerTy() && RetTy->isIntegerTy()) - Builder.CreateRet(Builder.CreatePtrToInt(CI, RetTy)); - else - Builder.CreateRet(Builder.CreateBitCast(CI, RetTy)); + Builder.CreateRet(createCast(Builder, CI, NewG->getReturnType())); } NewG->copyAttributesFrom(G); @@ -829,6 +844,18 @@ bool MergeFunctions::insert(ComparableFunction &NewF) { const ComparableFunction &OldF = *Result.first; + // Don't merge tiny functions, since it can just end up making the function + // larger. + // FIXME: Should still merge them if they are unnamed_addr and produce an + // alias. + if (NewF.getFunc()->size() == 1) { + if (NewF.getFunc()->front().size() <= 2) { + DEBUG(dbgs() << NewF.getFunc()->getName() + << " is to small to bother merging\n"); + return false; + } + } + // Never thunk a strong function to a weak function. assert(!OldF.getFunc()->mayBeOverridden() || NewF.getFunc()->mayBeOverridden()); diff --git a/contrib/llvm/lib/Transforms/IPO/PassManagerBuilder.cpp b/contrib/llvm/lib/Transforms/IPO/PassManagerBuilder.cpp index 986c0b8..24c5018 100644 --- a/contrib/llvm/lib/Transforms/IPO/PassManagerBuilder.cpp +++ b/contrib/llvm/lib/Transforms/IPO/PassManagerBuilder.cpp @@ -29,15 +29,20 @@ using namespace llvm; static cl::opt<bool> -RunLoopVectorization("vectorize-loops", +RunLoopVectorization("vectorize-loops", cl::Hidden, cl::desc("Run the Loop vectorization passes")); static cl::opt<bool> -RunSLPVectorization("vectorize-slp", +LateVectorization("late-vectorize", cl::init(true), cl::Hidden, + cl::desc("Run the vectorization pasess late in the pass " + "pipeline (after the inliner)")); + +static cl::opt<bool> +RunSLPVectorization("vectorize-slp", cl::Hidden, cl::desc("Run the SLP vectorization passes")); static cl::opt<bool> -RunBBVectorization("vectorize-slp-aggressive", +RunBBVectorization("vectorize-slp-aggressive", cl::Hidden, cl::desc("Run the BB vectorization passes")); static cl::opt<bool> @@ -49,17 +54,22 @@ static cl::opt<bool> UseNewSROA("use-new-sroa", cl::init(true), cl::Hidden, cl::desc("Enable the new, experimental SROA pass")); +static cl::opt<bool> +RunLoopRerolling("reroll-loops", cl::Hidden, + cl::desc("Run the loop rerolling pass")); + PassManagerBuilder::PassManagerBuilder() { OptLevel = 2; SizeLevel = 0; LibraryInfo = 0; Inliner = 0; - DisableSimplifyLibCalls = false; DisableUnitAtATime = false; DisableUnrollLoops = false; BBVectorize = RunBBVectorization; SLPVectorize = RunSLPVectorization; LoopVectorize = RunLoopVectorization; + LateVectorize = LateVectorization; + RerollLoops = RunLoopRerolling; } PassManagerBuilder::~PassManagerBuilder() { @@ -174,8 +184,6 @@ void PassManagerBuilder::populateModulePassManager(PassManagerBase &MPM) { else MPM.add(createScalarReplAggregatesPass(-1, false)); MPM.add(createEarlyCSEPass()); // Catch trivial redundancies - if (!DisableSimplifyLibCalls) - MPM.add(createSimplifyLibCallsPass()); // Library Call Optimizations MPM.add(createJumpThreadingPass()); // Thread jumps. MPM.add(createCorrelatedValuePropagationPass()); // Propagate conditionals MPM.add(createCFGSimplificationPass()); // Merge & remove BBs @@ -192,8 +200,8 @@ void PassManagerBuilder::populateModulePassManager(PassManagerBase &MPM) { MPM.add(createLoopIdiomPass()); // Recognize idioms like memset. MPM.add(createLoopDeletionPass()); // Delete dead loops - if (LoopVectorize && OptLevel > 2) - MPM.add(createLoopVectorizePass()); + if (!LateVectorize && LoopVectorize) + MPM.add(createLoopVectorizePass(DisableUnrollLoops)); if (!DisableUnrollLoops) MPM.add(createLoopUnrollPass()); // Unroll small loops @@ -213,16 +221,18 @@ void PassManagerBuilder::populateModulePassManager(PassManagerBase &MPM) { addExtensionsToPM(EP_ScalarOptimizerLate, MPM); + if (RerollLoops) + MPM.add(createLoopRerollPass()); if (SLPVectorize) - MPM.add(createSLPVectorizerPass()); // Vectorize parallel scalar chains. + MPM.add(createSLPVectorizerPass()); // Vectorize parallel scalar chains. if (BBVectorize) { MPM.add(createBBVectorizePass()); MPM.add(createInstructionCombiningPass()); if (OptLevel > 1 && UseGVNAfterVectorization) - MPM.add(createGVNPass()); // Remove redundancies + MPM.add(createGVNPass()); // Remove redundancies else - MPM.add(createEarlyCSEPass()); // Catch trivial redundancies + MPM.add(createEarlyCSEPass()); // Catch trivial redundancies // BBVectorize may have significantly shortened a loop body; unroll again. if (!DisableUnrollLoops) @@ -230,9 +240,25 @@ void PassManagerBuilder::populateModulePassManager(PassManagerBase &MPM) { } MPM.add(createAggressiveDCEPass()); // Delete dead instructions - MPM.add(createCFGSimplificationPass()); // Merge & remove BBs + MPM.add(createCFGSimplificationPass()); // Merge & remove BBs MPM.add(createInstructionCombiningPass()); // Clean up after everything. + // As an experimental mode, run any vectorization passes in a separate + // pipeline from the CGSCC pass manager that runs iteratively with the + // inliner. + if (LateVectorize && LoopVectorize) { + // FIXME: This is a HACK! The inliner pass above implicitly creates a CGSCC + // pass manager that we are specifically trying to avoid. To prevent this + // we must insert a no-op module pass to reset the pass manager. + MPM.add(createBarrierNoopPass()); + + // Add the various vectorization passes and relevant cleanup passes for + // them since we are no longer in the middle of the main scalar pipeline. + MPM.add(createLoopVectorizePass(DisableUnrollLoops)); + MPM.add(createInstructionCombiningPass()); + MPM.add(createCFGSimplificationPass()); + } + if (!DisableUnitAtATime) { // FIXME: We shouldn't bother with this anymore. MPM.add(createStripDeadPrototypesPass()); // Get rid of dead prototypes @@ -257,11 +283,8 @@ void PassManagerBuilder::populateLTOPassManager(PassManagerBase &PM, // Now that composite has been compiled, scan through the module, looking // for a main function. If main is defined, mark all other functions // internal. - if (Internalize) { - std::vector<const char*> E; - E.push_back("main"); - PM.add(createInternalizePass(E)); - } + if (Internalize) + PM.add(createInternalizePass("main")); // Propagate constants at call sites into the functions they call. This // opens opportunities for globalopt (and inlining) by substituting function @@ -302,6 +325,7 @@ void PassManagerBuilder::populateLTOPassManager(PassManagerBase &PM, // The IPO passes may leave cruft around. Clean up after them. PM.add(createInstructionCombiningPass()); PM.add(createJumpThreadingPass()); + // Break up allocas if (UseNewSROA) PM.add(createSROAPass()); @@ -315,6 +339,7 @@ void PassManagerBuilder::populateLTOPassManager(PassManagerBase &PM, PM.add(createLICMPass()); // Hoist loop invariants. PM.add(createGVNPass(DisableGVNLoadPRE)); // Remove redundancies. PM.add(createMemCpyOptPass()); // Remove dead memcpys. + // Nuke dead stores. PM.add(createDeadStoreEliminationPass()); @@ -379,8 +404,7 @@ LLVMPassManagerBuilderSetDisableUnrollLoops(LLVMPassManagerBuilderRef PMB, void LLVMPassManagerBuilderSetDisableSimplifyLibCalls(LLVMPassManagerBuilderRef PMB, LLVMBool Value) { - PassManagerBuilder *Builder = unwrap(PMB); - Builder->DisableSimplifyLibCalls = Value; + // NOTE: The simplify-libcalls pass has been removed. } void diff --git a/contrib/llvm/lib/Transforms/IPO/PruneEH.cpp b/contrib/llvm/lib/Transforms/IPO/PruneEH.cpp index 73d9323..b160913 100644 --- a/contrib/llvm/lib/Transforms/IPO/PruneEH.cpp +++ b/contrib/llvm/lib/Transforms/IPO/PruneEH.cpp @@ -51,7 +51,7 @@ namespace { char PruneEH::ID = 0; INITIALIZE_PASS_BEGIN(PruneEH, "prune-eh", "Remove unused exception handling info", false, false) -INITIALIZE_AG_DEPENDENCY(CallGraph) +INITIALIZE_PASS_DEPENDENCY(CallGraph) INITIALIZE_PASS_END(PruneEH, "prune-eh", "Remove unused exception handling info", false, false) @@ -145,15 +145,13 @@ bool PruneEH::runOnSCC(CallGraphSCC &SCC) { NewAttributes.addAttribute(Attribute::NoReturn); Function *F = (*I)->getFunction(); - const AttributeSet &PAL = F->getAttributes(); - const AttributeSet &NPAL = - PAL.addAttributes(F->getContext(), AttributeSet::FunctionIndex, - AttributeSet::get(F->getContext(), - AttributeSet::FunctionIndex, - NewAttributes)); + const AttributeSet &PAL = F->getAttributes().getFnAttributes(); + const AttributeSet &NPAL = AttributeSet::get( + F->getContext(), AttributeSet::FunctionIndex, NewAttributes); + if (PAL != NPAL) { MadeChange = true; - F->setAttributes(NPAL); + F->addAttributes(AttributeSet::FunctionIndex, NPAL); } } diff --git a/contrib/llvm/lib/Transforms/IPO/StripSymbols.cpp b/contrib/llvm/lib/Transforms/IPO/StripSymbols.cpp index 3396f79..c4f5cfc 100644 --- a/contrib/llvm/lib/Transforms/IPO/StripSymbols.cpp +++ b/contrib/llvm/lib/Transforms/IPO/StripSymbols.cpp @@ -9,7 +9,7 @@ // // The StripSymbols transformation implements code stripping. Specifically, it // can delete: -// +// // * names for virtual registers // * symbols for internal globals and functions // * debug information @@ -39,7 +39,7 @@ namespace { bool OnlyDebugInfo; public: static char ID; // Pass identification, replacement for typeid - explicit StripSymbols(bool ODI = false) + explicit StripSymbols(bool ODI = false) : ModulePass(ID), OnlyDebugInfo(ODI) { initializeStripSymbolsPass(*PassRegistry::getPassRegistry()); } @@ -144,7 +144,7 @@ static void RemoveDeadConstant(Constant *C) { assert(C->use_empty() && "Constant is not dead!"); SmallPtrSet<Constant*, 4> Operands; for (unsigned i = 0, e = C->getNumOperands(); i != e; ++i) - if (OnlyUsedBy(C->getOperand(i), C)) + if (OnlyUsedBy(C->getOperand(i), C)) Operands.insert(cast<Constant>(C->getOperand(i))); if (GlobalVariable *GV = dyn_cast<GlobalVariable>(C)) { if (!GV->hasLocalLinkage()) return; // Don't delete non static globals. @@ -182,7 +182,7 @@ static void StripTypeNames(Module &M, bool PreserveDbgInfo) { for (unsigned i = 0, e = StructTypes.size(); i != e; ++i) { StructType *STy = StructTypes[i]; if (STy->isLiteral() || STy->getName().empty()) continue; - + if (PreserveDbgInfo && STy->getName().startswith("llvm.dbg")) continue; @@ -199,7 +199,7 @@ static void findUsedValues(GlobalVariable *LLVMUsed, ConstantArray *Inits = cast<ConstantArray>(LLVMUsed->getInitializer()); for (unsigned i = 0, e = Inits->getNumOperands(); i != e; ++i) - if (GlobalValue *GV = + if (GlobalValue *GV = dyn_cast<GlobalValue>(Inits->getOperand(i)->stripPointerCasts())) UsedValues.insert(GV); } @@ -217,71 +217,20 @@ static bool StripSymbolNames(Module &M, bool PreserveDbgInfo) { if (!PreserveDbgInfo || !I->getName().startswith("llvm.dbg")) I->setName(""); // Internal symbols can't participate in linkage } - + for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I) { if (I->hasLocalLinkage() && llvmUsedValues.count(I) == 0) if (!PreserveDbgInfo || !I->getName().startswith("llvm.dbg")) I->setName(""); // Internal symbols can't participate in linkage StripSymtab(I->getValueSymbolTable(), PreserveDbgInfo); } - + // Remove all names from types. StripTypeNames(M, PreserveDbgInfo); return true; } -// StripDebugInfo - Strip debug info in the module if it exists. -// To do this, we remove llvm.dbg.func.start, llvm.dbg.stoppoint, and -// llvm.dbg.region.end calls, and any globals they point to if now dead. -static bool StripDebugInfo(Module &M) { - - bool Changed = false; - - // Remove all of the calls to the debugger intrinsics, and remove them from - // the module. - if (Function *Declare = M.getFunction("llvm.dbg.declare")) { - while (!Declare->use_empty()) { - CallInst *CI = cast<CallInst>(Declare->use_back()); - CI->eraseFromParent(); - } - Declare->eraseFromParent(); - Changed = true; - } - - if (Function *DbgVal = M.getFunction("llvm.dbg.value")) { - while (!DbgVal->use_empty()) { - CallInst *CI = cast<CallInst>(DbgVal->use_back()); - CI->eraseFromParent(); - } - DbgVal->eraseFromParent(); - Changed = true; - } - - for (Module::named_metadata_iterator NMI = M.named_metadata_begin(), - NME = M.named_metadata_end(); NMI != NME;) { - NamedMDNode *NMD = NMI; - ++NMI; - if (NMD->getName().startswith("llvm.dbg.")) { - NMD->eraseFromParent(); - Changed = true; - } - } - - for (Module::iterator MI = M.begin(), ME = M.end(); MI != ME; ++MI) - for (Function::iterator FI = MI->begin(), FE = MI->end(); FI != FE; - ++FI) - for (BasicBlock::iterator BI = FI->begin(), BE = FI->end(); BI != BE; - ++BI) { - if (!BI->getDebugLoc().isUnknown()) { - Changed = true; - BI->setDebugLoc(DebugLoc()); - } - } - - return Changed; -} - bool StripSymbols::runOnModule(Module &M) { bool Changed = false; Changed |= StripDebugInfo(M); @@ -307,13 +256,13 @@ bool StripDebugDeclare::runOnModule(Module &M) { assert(CI->use_empty() && "llvm.dbg intrinsic should have void result"); CI->eraseFromParent(); if (Arg1->use_empty()) { - if (Constant *C = dyn_cast<Constant>(Arg1)) + if (Constant *C = dyn_cast<Constant>(Arg1)) DeadConstants.push_back(C); - else + else RecursivelyDeleteTriviallyDeadInstructions(Arg1); } if (Arg2->use_empty()) - if (Constant *C = dyn_cast<Constant>(Arg2)) + if (Constant *C = dyn_cast<Constant>(Arg2)) DeadConstants.push_back(C); } Declare->eraseFromParent(); @@ -332,81 +281,107 @@ bool StripDebugDeclare::runOnModule(Module &M) { return true; } -/// getRealLinkageName - If special LLVM prefix that is used to inform the asm -/// printer to not emit usual symbol prefix before the symbol name is used then -/// return linkage name after skipping this special LLVM prefix. -static StringRef getRealLinkageName(StringRef LinkageName) { - char One = '\1'; - if (LinkageName.startswith(StringRef(&One, 1))) - return LinkageName.substr(1); - return LinkageName; -} - +/// Remove any debug info for global variables/functions in the given module for +/// which said global variable/function no longer exists (i.e. is null). +/// +/// Debugging information is encoded in llvm IR using metadata. This is designed +/// such a way that debug info for symbols preserved even if symbols are +/// optimized away by the optimizer. This special pass removes debug info for +/// such symbols. bool StripDeadDebugInfo::runOnModule(Module &M) { bool Changed = false; - // Debugging infomration is encoded in llvm IR using metadata. This is designed - // such a way that debug info for symbols preserved even if symbols are - // optimized away by the optimizer. This special pass removes debug info for - // such symbols. - - // llvm.dbg.gv keeps track of debug info for global variables. - if (NamedMDNode *NMD = M.getNamedMetadata("llvm.dbg.gv")) { - SmallVector<MDNode *, 8> MDs; - for (unsigned i = 0, e = NMD->getNumOperands(); i != e; ++i) - if (DIGlobalVariable(NMD->getOperand(i)).Verify()) - MDs.push_back(NMD->getOperand(i)); - else - Changed = true; - NMD->eraseFromParent(); - NMD = NULL; - - for (SmallVector<MDNode *, 8>::iterator I = MDs.begin(), - E = MDs.end(); I != E; ++I) { - GlobalVariable *GV = DIGlobalVariable(*I).getGlobal(); - if (GV && M.getGlobalVariable(GV->getName(), true)) { - if (!NMD) - NMD = M.getOrInsertNamedMetadata("llvm.dbg.gv"); - NMD->addOperand(*I); - } + LLVMContext &C = M.getContext(); + + // Find all debug info in F. This is actually overkill in terms of what we + // want to do, but we want to try and be as resilient as possible in the face + // of potential debug info changes by using the formal interfaces given to us + // as much as possible. + DebugInfoFinder F; + F.processModule(M); + + // For each compile unit, find the live set of global variables/functions and + // replace the current list of potentially dead global variables/functions + // with the live list. + SmallVector<Value *, 64> LiveGlobalVariables; + SmallVector<Value *, 64> LiveSubprograms; + DenseSet<const MDNode *> VisitedSet; + + for (DebugInfoFinder::iterator CI = F.compile_unit_begin(), + CE = F.compile_unit_end(); CI != CE; ++CI) { + // Create our compile unit. + DICompileUnit DIC(*CI); + assert(DIC.Verify() && "DIC must verify as a DICompileUnit."); + + // Create our live subprogram list. + DIArray SPs = DIC.getSubprograms(); + bool SubprogramChange = false; + for (unsigned i = 0, e = SPs.getNumElements(); i != e; ++i) { + DISubprogram DISP(SPs.getElement(i)); + assert(DISP.Verify() && "DISP must verify as a DISubprogram."); + + // Make sure we visit each subprogram only once. + if (!VisitedSet.insert(DISP).second) + continue; + + // If the function referenced by DISP is not null, the function is live. + if (DISP.getFunction()) + LiveSubprograms.push_back(DISP); else - Changed = true; + SubprogramChange = true; } - } - // llvm.dbg.sp keeps track of debug info for subprograms. - if (NamedMDNode *NMD = M.getNamedMetadata("llvm.dbg.sp")) { - SmallVector<MDNode *, 8> MDs; - for (unsigned i = 0, e = NMD->getNumOperands(); i != e; ++i) - if (DISubprogram(NMD->getOperand(i)).Verify()) - MDs.push_back(NMD->getOperand(i)); + // Create our live global variable list. + DIArray GVs = DIC.getGlobalVariables(); + bool GlobalVariableChange = false; + for (unsigned i = 0, e = GVs.getNumElements(); i != e; ++i) { + DIGlobalVariable DIG(GVs.getElement(i)); + assert(DIG.Verify() && "DIG must verify as DIGlobalVariable."); + + // Make sure we only visit each global variable only once. + if (!VisitedSet.insert(DIG).second) + continue; + + // If the global variable referenced by DIG is not null, the global + // variable is live. + if (DIG.getGlobal()) + LiveGlobalVariables.push_back(DIG); else - Changed = true; - NMD->eraseFromParent(); - NMD = NULL; - - for (SmallVector<MDNode *, 8>::iterator I = MDs.begin(), - E = MDs.end(); I != E; ++I) { - bool FnIsLive = false; - if (Function *F = DISubprogram(*I).getFunction()) - if (M.getFunction(F->getName())) - FnIsLive = true; - if (FnIsLive) { - if (!NMD) - NMD = M.getOrInsertNamedMetadata("llvm.dbg.sp"); - NMD->addOperand(*I); - } else { - // Remove llvm.dbg.lv.fnname named mdnode which may have been used - // to hold debug info for dead function's local variables. - StringRef FName = DISubprogram(*I).getLinkageName(); - if (FName.empty()) - FName = DISubprogram(*I).getName(); - if (NamedMDNode *LVNMD = - M.getNamedMetadata(Twine("llvm.dbg.lv.", - getRealLinkageName(FName)))) - LVNMD->eraseFromParent(); - } + GlobalVariableChange = true; + } + + // If we found dead subprograms or global variables, replace the current + // subprogram list/global variable list with our new live subprogram/global + // variable list. + if (SubprogramChange) { + // Make sure that 9 is still the index of the subprograms. This is to make + // sure that an assert is hit if the location of the subprogram array + // changes. This is just to make sure that this is updated if such an + // event occurs. + assert(DIC->getNumOperands() >= 10 && + SPs == DIC->getOperand(9) && + "DICompileUnits is expected to store Subprograms in operand " + "9."); + DIC->replaceOperandWith(9, MDNode::get(C, LiveSubprograms)); + Changed = true; } + + if (GlobalVariableChange) { + // Make sure that 10 is still the index of global variables. This is to + // make sure that an assert is hit if the location of the subprogram array + // changes. This is just to make sure that this index is updated if such + // an event occurs. + assert(DIC->getNumOperands() >= 11 && + GVs == DIC->getOperand(10) && + "DICompileUnits is expected to store Global Variables in operand " + "10."); + DIC->replaceOperandWith(10, MDNode::get(C, LiveGlobalVariables)); + Changed = true; + } + + // Reset lists for the next iteration. + LiveSubprograms.clear(); + LiveGlobalVariables.clear(); } return Changed; |