llvm/lib/Analysis/ScopedNoAliasAA.cpp
Chandler Carruth cf88e9244e [AA] Hoist the logic to reformulate various AA queries in terms of other
parts of the AA interface out of the base class of every single AA
result object.

Because this logic reformulates the query in terms of some other aspect
of the API, it would easily cause O(n^2) query patterns in alias
analysis. These could in turn be magnified further based on the number
of call arguments, and then further based on the number of AA queries
made for a particular call. This ended up causing problems for Rust that
were actually noticable enough to get a bug (PR26564) and probably other
places as well.

When originally re-working the AA infrastructure, the desire was to
regularize the pattern of refinement without losing any generality.
While I think it was successful, that is clearly proving to be too
costly. And the cost is needless: we gain no actual improvement for this
generality of making a direct query to tbaa actually be able to
re-use some other alias analysis's refinement logic for one of the other
APIs, or some such. In short, this is entirely wasted work.

To the extent possible, delegation to other API surfaces should be done
at the aggregation layer so that we can avoid re-walking the
aggregation. In fact, this significantly simplifies the logic as we no
longer need to smuggle the aggregation layer into each alias analysis
(or the TargetLibraryInfo into each alias analysis just so we can form
argument memory locations!).

However, we also have some delegation logic inside of BasicAA and some
of it even makes sense. When the delegation logic is baking in specific
knowledge of aliasing properties of the LLVM IR, as opposed to simply
reformulating the query to utilize a different alias analysis interface
entry point, it makes a lot of sense to restrict that logic to
a different layer such as BasicAA. So one aspect of the delegation that
was in every AA base class is that when we don't have operand bundles,
we re-use function AA results as a fallback for callsite alias results.
This relies on the IR properties of calls and functions w.r.t. aliasing,
and so seems a better fit to BasicAA. I've lifted the logic up to that
point where it seems to be a natural fit. This still does a bit of
redundant work (we query function attributes twice, once via the
callsite and once via the function AA query) but it is *exactly* twice
here, no more.

The end result is that all of the delegation logic is hoisted out of the
base class and into either the aggregation layer when it is a pure
retargeting to a different API surface, or into BasicAA when it relies
on the IR's aliasing properties. This should fix the quadratic query
pattern reported in PR26564, although I don't have a stand-alone test
case to reproduce it.

It also seems general goodness. Now the numerous AAs that don't need
target library info don't carry it around and depend on it. I think
I can even rip out the general access to the aggregation layer and only
expose that in BasicAA as it is the only place where we re-query in that
manner.

However, this is a non-trivial change to the AA infrastructure so I want
to get some additional eyes on this before it lands. Sadly, it can't
wait long because we should really cherry pick this into 3.8 if we're
going to go this route.

Differential Revision: http://reviews.llvm.org/D17329

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@262490 91177308-0d34-0410-b5e6-96231b3b80d8
2016-03-02 15:56:53 +00:00

205 lines
7.0 KiB
C++

//===- ScopedNoAliasAA.cpp - Scoped No-Alias Alias Analysis ---------------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file defines the ScopedNoAlias alias-analysis pass, which implements
// metadata-based scoped no-alias support.
//
// Alias-analysis scopes are defined by an id (which can be a string or some
// other metadata node), a domain node, and an optional descriptive string.
// A domain is defined by an id (which can be a string or some other metadata
// node), and an optional descriptive string.
//
// !dom0 = metadata !{ metadata !"domain of foo()" }
// !scope1 = metadata !{ metadata !scope1, metadata !dom0, metadata !"scope 1" }
// !scope2 = metadata !{ metadata !scope2, metadata !dom0, metadata !"scope 2" }
//
// Loads and stores can be tagged with an alias-analysis scope, and also, with
// a noalias tag for a specific scope:
//
// ... = load %ptr1, !alias.scope !{ !scope1 }
// ... = load %ptr2, !alias.scope !{ !scope1, !scope2 }, !noalias !{ !scope1 }
//
// When evaluating an aliasing query, if one of the instructions is associated
// has a set of noalias scopes in some domain that is a superset of the alias
// scopes in that domain of some other instruction, then the two memory
// accesses are assumed not to alias.
//
//===----------------------------------------------------------------------===//
#include "llvm/Analysis/ScopedNoAliasAA.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/LLVMContext.h"
#include "llvm/IR/Metadata.h"
#include "llvm/IR/Module.h"
#include "llvm/Pass.h"
#include "llvm/Support/CommandLine.h"
using namespace llvm;
// A handy option for disabling scoped no-alias functionality. The same effect
// can also be achieved by stripping the associated metadata tags from IR, but
// this option is sometimes more convenient.
static cl::opt<bool> EnableScopedNoAlias("enable-scoped-noalias",
cl::init(true));
namespace {
/// This is a simple wrapper around an MDNode which provides a higher-level
/// interface by hiding the details of how alias analysis information is encoded
/// in its operands.
class AliasScopeNode {
const MDNode *Node;
public:
AliasScopeNode() : Node(nullptr) {}
explicit AliasScopeNode(const MDNode *N) : Node(N) {}
/// Get the MDNode for this AliasScopeNode.
const MDNode *getNode() const { return Node; }
/// Get the MDNode for this AliasScopeNode's domain.
const MDNode *getDomain() const {
if (Node->getNumOperands() < 2)
return nullptr;
return dyn_cast_or_null<MDNode>(Node->getOperand(1));
}
};
} // end of anonymous namespace
AliasResult ScopedNoAliasAAResult::alias(const MemoryLocation &LocA,
const MemoryLocation &LocB) {
if (!EnableScopedNoAlias)
return AAResultBase::alias(LocA, LocB);
// Get the attached MDNodes.
const MDNode *AScopes = LocA.AATags.Scope, *BScopes = LocB.AATags.Scope;
const MDNode *ANoAlias = LocA.AATags.NoAlias, *BNoAlias = LocB.AATags.NoAlias;
if (!mayAliasInScopes(AScopes, BNoAlias))
return NoAlias;
if (!mayAliasInScopes(BScopes, ANoAlias))
return NoAlias;
// If they may alias, chain to the next AliasAnalysis.
return AAResultBase::alias(LocA, LocB);
}
ModRefInfo ScopedNoAliasAAResult::getModRefInfo(ImmutableCallSite CS,
const MemoryLocation &Loc) {
if (!EnableScopedNoAlias)
return AAResultBase::getModRefInfo(CS, Loc);
if (!mayAliasInScopes(Loc.AATags.Scope, CS.getInstruction()->getMetadata(
LLVMContext::MD_noalias)))
return MRI_NoModRef;
if (!mayAliasInScopes(
CS.getInstruction()->getMetadata(LLVMContext::MD_alias_scope),
Loc.AATags.NoAlias))
return MRI_NoModRef;
return AAResultBase::getModRefInfo(CS, Loc);
}
ModRefInfo ScopedNoAliasAAResult::getModRefInfo(ImmutableCallSite CS1,
ImmutableCallSite CS2) {
if (!EnableScopedNoAlias)
return AAResultBase::getModRefInfo(CS1, CS2);
if (!mayAliasInScopes(
CS1.getInstruction()->getMetadata(LLVMContext::MD_alias_scope),
CS2.getInstruction()->getMetadata(LLVMContext::MD_noalias)))
return MRI_NoModRef;
if (!mayAliasInScopes(
CS2.getInstruction()->getMetadata(LLVMContext::MD_alias_scope),
CS1.getInstruction()->getMetadata(LLVMContext::MD_noalias)))
return MRI_NoModRef;
return AAResultBase::getModRefInfo(CS1, CS2);
}
void ScopedNoAliasAAResult::collectMDInDomain(
const MDNode *List, const MDNode *Domain,
SmallPtrSetImpl<const MDNode *> &Nodes) const {
for (const MDOperand &MDOp : List->operands())
if (const MDNode *MD = dyn_cast<MDNode>(MDOp))
if (AliasScopeNode(MD).getDomain() == Domain)
Nodes.insert(MD);
}
bool ScopedNoAliasAAResult::mayAliasInScopes(const MDNode *Scopes,
const MDNode *NoAlias) const {
if (!Scopes || !NoAlias)
return true;
// Collect the set of scope domains relevant to the noalias scopes.
SmallPtrSet<const MDNode *, 16> Domains;
for (const MDOperand &MDOp : NoAlias->operands())
if (const MDNode *NAMD = dyn_cast<MDNode>(MDOp))
if (const MDNode *Domain = AliasScopeNode(NAMD).getDomain())
Domains.insert(Domain);
// We alias unless, for some domain, the set of noalias scopes in that domain
// is a superset of the set of alias scopes in that domain.
for (const MDNode *Domain : Domains) {
SmallPtrSet<const MDNode *, 16> NANodes, ScopeNodes;
collectMDInDomain(NoAlias, Domain, NANodes);
collectMDInDomain(Scopes, Domain, ScopeNodes);
if (!ScopeNodes.size())
continue;
// To not alias, all of the nodes in ScopeNodes must be in NANodes.
bool FoundAll = true;
for (const MDNode *SMD : ScopeNodes)
if (!NANodes.count(SMD)) {
FoundAll = false;
break;
}
if (FoundAll)
return false;
}
return true;
}
ScopedNoAliasAAResult ScopedNoAliasAA::run(Function &F,
AnalysisManager<Function> *AM) {
return ScopedNoAliasAAResult();
}
char ScopedNoAliasAAWrapperPass::ID = 0;
INITIALIZE_PASS(ScopedNoAliasAAWrapperPass, "scoped-noalias",
"Scoped NoAlias Alias Analysis", false, true)
ImmutablePass *llvm::createScopedNoAliasAAWrapperPass() {
return new ScopedNoAliasAAWrapperPass();
}
ScopedNoAliasAAWrapperPass::ScopedNoAliasAAWrapperPass() : ImmutablePass(ID) {
initializeScopedNoAliasAAWrapperPassPass(*PassRegistry::getPassRegistry());
}
bool ScopedNoAliasAAWrapperPass::doInitialization(Module &M) {
Result.reset(new ScopedNoAliasAAResult());
return false;
}
bool ScopedNoAliasAAWrapperPass::doFinalization(Module &M) {
Result.reset();
return false;
}
void ScopedNoAliasAAWrapperPass::getAnalysisUsage(AnalysisUsage &AU) const {
AU.setPreservesAll();
}