1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-19 19:12:56 +02:00
llvm-mirror/lib/Bitcode/Writer/ValueEnumerator.cpp

991 lines
32 KiB
C++
Raw Normal View History

//===-- ValueEnumerator.cpp - Number values and types for bitcode writer --===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file implements the ValueEnumerator class.
//
//===----------------------------------------------------------------------===//
#include "ValueEnumerator.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/DebugInfoMetadata.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/Module.h"
#include "llvm/IR/UseListOrder.h"
#include "llvm/IR/ValueSymbolTable.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
#include <algorithm>
using namespace llvm;
namespace {
struct OrderMap {
DenseMap<const Value *, std::pair<unsigned, bool>> IDs;
unsigned LastGlobalConstantID;
unsigned LastGlobalValueID;
OrderMap() : LastGlobalConstantID(0), LastGlobalValueID(0) {}
bool isGlobalConstant(unsigned ID) const {
return ID <= LastGlobalConstantID;
}
bool isGlobalValue(unsigned ID) const {
return ID <= LastGlobalValueID && !isGlobalConstant(ID);
}
unsigned size() const { return IDs.size(); }
std::pair<unsigned, bool> &operator[](const Value *V) { return IDs[V]; }
std::pair<unsigned, bool> lookup(const Value *V) const {
return IDs.lookup(V);
}
void index(const Value *V) {
// Explicitly sequence get-size and insert-value operations to avoid UB.
unsigned ID = IDs.size() + 1;
IDs[V].first = ID;
}
};
}
static void orderValue(const Value *V, OrderMap &OM) {
if (OM.lookup(V).first)
return;
if (const Constant *C = dyn_cast<Constant>(V))
if (C->getNumOperands() && !isa<GlobalValue>(C))
for (const Value *Op : C->operands())
if (!isa<BasicBlock>(Op) && !isa<GlobalValue>(Op))
orderValue(Op, OM);
// Note: we cannot cache this lookup above, since inserting into the map
// changes the map's size, and thus affects the other IDs.
OM.index(V);
}
static OrderMap orderModule(const Module &M) {
// This needs to match the order used by ValueEnumerator::ValueEnumerator()
// and ValueEnumerator::incorporateFunction().
OrderMap OM;
// In the reader, initializers of GlobalValues are set *after* all the
// globals have been read. Rather than awkwardly modeling this behaviour
// directly in predictValueUseListOrderImpl(), just assign IDs to
// initializers of GlobalValues before GlobalValues themselves to model this
// implicitly.
for (const GlobalVariable &G : M.globals())
if (G.hasInitializer())
if (!isa<GlobalValue>(G.getInitializer()))
orderValue(G.getInitializer(), OM);
for (const GlobalAlias &A : M.aliases())
if (!isa<GlobalValue>(A.getAliasee()))
orderValue(A.getAliasee(), OM);
for (const GlobalIFunc &I : M.ifuncs())
if (!isa<GlobalValue>(I.getResolver()))
orderValue(I.getResolver(), OM);
Prologue support Patch by Ben Gamari! This redefines the `prefix` attribute introduced previously and introduces a `prologue` attribute. There are a two primary usecases that these attributes aim to serve, 1. Function prologue sigils 2. Function hot-patching: Enable the user to insert `nop` operations at the beginning of the function which can later be safely replaced with a call to some instrumentation facility 3. Runtime metadata: Allow a compiler to insert data for use by the runtime during execution. GHC is one example of a compiler that needs this functionality for its tables-next-to-code functionality. Previously `prefix` served cases (1) and (2) quite well by allowing the user to introduce arbitrary data at the entrypoint but before the function body. Case (3), however, was poorly handled by this approach as it required that prefix data was valid executable code. Here we redefine the notion of prefix data to instead be data which occurs immediately before the function entrypoint (i.e. the symbol address). Since prefix data now occurs before the function entrypoint, there is no need for the data to be valid code. The previous notion of prefix data now goes under the name "prologue data" to emphasize its duality with the function epilogue. The intention here is to handle cases (1) and (2) with prologue data and case (3) with prefix data. References ---------- This idea arose out of discussions[1] with Reid Kleckner in response to a proposal to introduce the notion of symbol offsets to enable handling of case (3). [1] http://lists.cs.uiuc.edu/pipermail/llvmdev/2014-May/073235.html Test Plan: testsuite Differential Revision: http://reviews.llvm.org/D6454 llvm-svn: 223189
2014-12-03 03:08:38 +01:00
for (const Function &F : M) {
for (const Use &U : F.operands())
if (!isa<GlobalValue>(U.get()))
orderValue(U.get(), OM);
Prologue support Patch by Ben Gamari! This redefines the `prefix` attribute introduced previously and introduces a `prologue` attribute. There are a two primary usecases that these attributes aim to serve, 1. Function prologue sigils 2. Function hot-patching: Enable the user to insert `nop` operations at the beginning of the function which can later be safely replaced with a call to some instrumentation facility 3. Runtime metadata: Allow a compiler to insert data for use by the runtime during execution. GHC is one example of a compiler that needs this functionality for its tables-next-to-code functionality. Previously `prefix` served cases (1) and (2) quite well by allowing the user to introduce arbitrary data at the entrypoint but before the function body. Case (3), however, was poorly handled by this approach as it required that prefix data was valid executable code. Here we redefine the notion of prefix data to instead be data which occurs immediately before the function entrypoint (i.e. the symbol address). Since prefix data now occurs before the function entrypoint, there is no need for the data to be valid code. The previous notion of prefix data now goes under the name "prologue data" to emphasize its duality with the function epilogue. The intention here is to handle cases (1) and (2) with prologue data and case (3) with prefix data. References ---------- This idea arose out of discussions[1] with Reid Kleckner in response to a proposal to introduce the notion of symbol offsets to enable handling of case (3). [1] http://lists.cs.uiuc.edu/pipermail/llvmdev/2014-May/073235.html Test Plan: testsuite Differential Revision: http://reviews.llvm.org/D6454 llvm-svn: 223189
2014-12-03 03:08:38 +01:00
}
OM.LastGlobalConstantID = OM.size();
// Initializers of GlobalValues are processed in
// BitcodeReader::ResolveGlobalAndAliasInits(). Match the order there rather
// than ValueEnumerator, and match the code in predictValueUseListOrderImpl()
// by giving IDs in reverse order.
//
// Since GlobalValues never reference each other directly (just through
// initializers), their relative IDs only matter for determining order of
// uses in their initializers.
for (const Function &F : M)
orderValue(&F, OM);
for (const GlobalAlias &A : M.aliases())
orderValue(&A, OM);
for (const GlobalIFunc &I : M.ifuncs())
orderValue(&I, OM);
for (const GlobalVariable &G : M.globals())
orderValue(&G, OM);
OM.LastGlobalValueID = OM.size();
for (const Function &F : M) {
if (F.isDeclaration())
continue;
// Here we need to match the union of ValueEnumerator::incorporateFunction()
// and WriteFunction(). Basic blocks are implicitly declared before
// anything else (by declaring their size).
for (const BasicBlock &BB : F)
orderValue(&BB, OM);
for (const Argument &A : F.args())
orderValue(&A, OM);
for (const BasicBlock &BB : F)
for (const Instruction &I : BB)
for (const Value *Op : I.operands())
if ((isa<Constant>(*Op) && !isa<GlobalValue>(*Op)) ||
isa<InlineAsm>(*Op))
orderValue(Op, OM);
for (const BasicBlock &BB : F)
for (const Instruction &I : BB)
orderValue(&I, OM);
}
return OM;
}
static void predictValueUseListOrderImpl(const Value *V, const Function *F,
unsigned ID, const OrderMap &OM,
UseListOrderStack &Stack) {
// Predict use-list order for this one.
typedef std::pair<const Use *, unsigned> Entry;
SmallVector<Entry, 64> List;
for (const Use &U : V->uses())
// Check if this user will be serialized.
if (OM.lookup(U.getUser()).first)
List.push_back(std::make_pair(&U, List.size()));
if (List.size() < 2)
// We may have lost some users.
return;
bool IsGlobalValue = OM.isGlobalValue(ID);
std::sort(List.begin(), List.end(), [&](const Entry &L, const Entry &R) {
const Use *LU = L.first;
const Use *RU = R.first;
if (LU == RU)
return false;
auto LID = OM.lookup(LU->getUser()).first;
auto RID = OM.lookup(RU->getUser()).first;
// Global values are processed in reverse order.
//
// Moreover, initializers of GlobalValues are set *after* all the globals
// have been read (despite having earlier IDs). Rather than awkwardly
// modeling this behaviour here, orderModule() has assigned IDs to
// initializers of GlobalValues before GlobalValues themselves.
if (OM.isGlobalValue(LID) && OM.isGlobalValue(RID))
return LID < RID;
// If ID is 4, then expect: 7 6 5 1 2 3.
if (LID < RID) {
if (RID <= ID)
if (!IsGlobalValue) // GlobalValue uses don't get reversed.
return true;
return false;
}
if (RID < LID) {
if (LID <= ID)
if (!IsGlobalValue) // GlobalValue uses don't get reversed.
return false;
return true;
}
// LID and RID are equal, so we have different operands of the same user.
// Assume operands are added in order for all instructions.
if (LID <= ID)
if (!IsGlobalValue) // GlobalValue uses don't get reversed.
return LU->getOperandNo() < RU->getOperandNo();
return LU->getOperandNo() > RU->getOperandNo();
});
if (std::is_sorted(
List.begin(), List.end(),
[](const Entry &L, const Entry &R) { return L.second < R.second; }))
// Order is already correct.
return;
// Store the shuffle.
Stack.emplace_back(V, F, List.size());
assert(List.size() == Stack.back().Shuffle.size() && "Wrong size");
for (size_t I = 0, E = List.size(); I != E; ++I)
Stack.back().Shuffle[I] = List[I].second;
}
static void predictValueUseListOrder(const Value *V, const Function *F,
OrderMap &OM, UseListOrderStack &Stack) {
auto &IDPair = OM[V];
assert(IDPair.first && "Unmapped value");
if (IDPair.second)
// Already predicted.
return;
// Do the actual prediction.
IDPair.second = true;
if (!V->use_empty() && std::next(V->use_begin()) != V->use_end())
predictValueUseListOrderImpl(V, F, IDPair.first, OM, Stack);
// Recursive descent into constants.
if (const Constant *C = dyn_cast<Constant>(V))
if (C->getNumOperands()) // Visit GlobalValues.
for (const Value *Op : C->operands())
if (isa<Constant>(Op)) // Visit GlobalValues.
predictValueUseListOrder(Op, F, OM, Stack);
}
static UseListOrderStack predictUseListOrder(const Module &M) {
OrderMap OM = orderModule(M);
// Use-list orders need to be serialized after all the users have been added
// to a value, or else the shuffles will be incomplete. Store them per
// function in a stack.
//
// Aside from function order, the order of values doesn't matter much here.
UseListOrderStack Stack;
// We want to visit the functions backward now so we can list function-local
// constants in the last Function they're used in. Module-level constants
// have already been visited above.
for (auto I = M.rbegin(), E = M.rend(); I != E; ++I) {
const Function &F = *I;
if (F.isDeclaration())
continue;
for (const BasicBlock &BB : F)
predictValueUseListOrder(&BB, &F, OM, Stack);
for (const Argument &A : F.args())
predictValueUseListOrder(&A, &F, OM, Stack);
for (const BasicBlock &BB : F)
for (const Instruction &I : BB)
for (const Value *Op : I.operands())
if (isa<Constant>(*Op) || isa<InlineAsm>(*Op)) // Visit GlobalValues.
predictValueUseListOrder(Op, &F, OM, Stack);
for (const BasicBlock &BB : F)
for (const Instruction &I : BB)
predictValueUseListOrder(&I, &F, OM, Stack);
}
// Visit globals last, since the module-level use-list block will be seen
// before the function bodies are processed.
for (const GlobalVariable &G : M.globals())
predictValueUseListOrder(&G, nullptr, OM, Stack);
for (const Function &F : M)
predictValueUseListOrder(&F, nullptr, OM, Stack);
for (const GlobalAlias &A : M.aliases())
predictValueUseListOrder(&A, nullptr, OM, Stack);
for (const GlobalIFunc &I : M.ifuncs())
predictValueUseListOrder(&I, nullptr, OM, Stack);
for (const GlobalVariable &G : M.globals())
if (G.hasInitializer())
predictValueUseListOrder(G.getInitializer(), nullptr, OM, Stack);
for (const GlobalAlias &A : M.aliases())
predictValueUseListOrder(A.getAliasee(), nullptr, OM, Stack);
for (const GlobalIFunc &I : M.ifuncs())
predictValueUseListOrder(I.getResolver(), nullptr, OM, Stack);
Prologue support Patch by Ben Gamari! This redefines the `prefix` attribute introduced previously and introduces a `prologue` attribute. There are a two primary usecases that these attributes aim to serve, 1. Function prologue sigils 2. Function hot-patching: Enable the user to insert `nop` operations at the beginning of the function which can later be safely replaced with a call to some instrumentation facility 3. Runtime metadata: Allow a compiler to insert data for use by the runtime during execution. GHC is one example of a compiler that needs this functionality for its tables-next-to-code functionality. Previously `prefix` served cases (1) and (2) quite well by allowing the user to introduce arbitrary data at the entrypoint but before the function body. Case (3), however, was poorly handled by this approach as it required that prefix data was valid executable code. Here we redefine the notion of prefix data to instead be data which occurs immediately before the function entrypoint (i.e. the symbol address). Since prefix data now occurs before the function entrypoint, there is no need for the data to be valid code. The previous notion of prefix data now goes under the name "prologue data" to emphasize its duality with the function epilogue. The intention here is to handle cases (1) and (2) with prologue data and case (3) with prefix data. References ---------- This idea arose out of discussions[1] with Reid Kleckner in response to a proposal to introduce the notion of symbol offsets to enable handling of case (3). [1] http://lists.cs.uiuc.edu/pipermail/llvmdev/2014-May/073235.html Test Plan: testsuite Differential Revision: http://reviews.llvm.org/D6454 llvm-svn: 223189
2014-12-03 03:08:38 +01:00
for (const Function &F : M) {
for (const Use &U : F.operands())
predictValueUseListOrder(U.get(), nullptr, OM, Stack);
Prologue support Patch by Ben Gamari! This redefines the `prefix` attribute introduced previously and introduces a `prologue` attribute. There are a two primary usecases that these attributes aim to serve, 1. Function prologue sigils 2. Function hot-patching: Enable the user to insert `nop` operations at the beginning of the function which can later be safely replaced with a call to some instrumentation facility 3. Runtime metadata: Allow a compiler to insert data for use by the runtime during execution. GHC is one example of a compiler that needs this functionality for its tables-next-to-code functionality. Previously `prefix` served cases (1) and (2) quite well by allowing the user to introduce arbitrary data at the entrypoint but before the function body. Case (3), however, was poorly handled by this approach as it required that prefix data was valid executable code. Here we redefine the notion of prefix data to instead be data which occurs immediately before the function entrypoint (i.e. the symbol address). Since prefix data now occurs before the function entrypoint, there is no need for the data to be valid code. The previous notion of prefix data now goes under the name "prologue data" to emphasize its duality with the function epilogue. The intention here is to handle cases (1) and (2) with prologue data and case (3) with prefix data. References ---------- This idea arose out of discussions[1] with Reid Kleckner in response to a proposal to introduce the notion of symbol offsets to enable handling of case (3). [1] http://lists.cs.uiuc.edu/pipermail/llvmdev/2014-May/073235.html Test Plan: testsuite Differential Revision: http://reviews.llvm.org/D6454 llvm-svn: 223189
2014-12-03 03:08:38 +01:00
}
return Stack;
}
static bool isIntOrIntVectorValue(const std::pair<const Value*, unsigned> &V) {
return V.first->getType()->isIntOrIntVectorTy();
}
ValueEnumerator::ValueEnumerator(const Module &M,
bool ShouldPreserveUseListOrder)
: ShouldPreserveUseListOrder(ShouldPreserveUseListOrder) {
if (ShouldPreserveUseListOrder)
UseListOrders = predictUseListOrder(M);
// Enumerate the global variables.
for (const GlobalVariable &GV : M.globals())
EnumerateValue(&GV);
// Enumerate the functions.
for (const Function & F : M) {
EnumerateValue(&F);
EnumerateAttributes(F.getAttributes());
}
// Enumerate the aliases.
for (const GlobalAlias &GA : M.aliases())
EnumerateValue(&GA);
// Enumerate the ifuncs.
for (const GlobalIFunc &GIF : M.ifuncs())
EnumerateValue(&GIF);
// Remember what is the cutoff between globalvalue's and other constants.
unsigned FirstConstant = Values.size();
// Enumerate the global variable initializers.
for (const GlobalVariable &GV : M.globals())
if (GV.hasInitializer())
EnumerateValue(GV.getInitializer());
// Enumerate the aliasees.
for (const GlobalAlias &GA : M.aliases())
EnumerateValue(GA.getAliasee());
// Enumerate the ifunc resolvers.
for (const GlobalIFunc &GIF : M.ifuncs())
EnumerateValue(GIF.getResolver());
// Enumerate any optional Function data.
for (const Function &F : M)
for (const Use &U : F.operands())
EnumerateValue(U.get());
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
// Enumerate the metadata type.
//
// TODO: Move this to ValueEnumerator::EnumerateOperandType() once bitcode
// only encodes the metadata type when it's used as a value.
EnumerateType(Type::getMetadataTy(M.getContext()));
2013-04-01 04:28:07 +02:00
// Insert constants and metadata that are named at module level into the slot
// pool so that the module symbol table can refer to them...
EnumerateValueSymbolTable(M.getValueSymbolTable());
EnumerateNamedMetadata(M);
SmallVector<std::pair<unsigned, MDNode *>, 8> MDs;
2009-12-31 01:51:46 +01:00
// Enumerate types used by function bodies and argument lists.
for (const Function &F : M) {
for (const Argument &A : F.args())
EnumerateType(A.getType());
// Enumerate metadata attached to this function.
F.getAllMetadata(MDs);
for (const auto &I : MDs)
2016-04-02 17:22:57 +02:00
EnumerateMetadata(&F, I.second);
for (const BasicBlock &BB : F)
for (const Instruction &I : BB) {
for (const Use &Op : I.operands()) {
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
auto *MD = dyn_cast<MetadataAsValue>(&Op);
if (!MD) {
EnumerateOperandType(Op);
continue;
}
// Local metadata is enumerated during function-incorporation.
if (isa<LocalAsMetadata>(MD->getMetadata()))
continue;
2016-04-02 17:22:57 +02:00
EnumerateMetadata(&F, MD->getMetadata());
}
EnumerateType(I.getType());
if (const CallInst *CI = dyn_cast<CallInst>(&I))
EnumerateAttributes(CI->getAttributes());
else if (const InvokeInst *II = dyn_cast<InvokeInst>(&I))
EnumerateAttributes(II->getAttributes());
// Enumerate metadata attached with this instruction.
MDs.clear();
I.getAllMetadataOtherThanDebugLoc(MDs);
for (unsigned i = 0, e = MDs.size(); i != e; ++i)
2016-04-02 17:22:57 +02:00
EnumerateMetadata(&F, MDs[i].second);
// Don't enumerate the location directly -- it has a special record
// type -- but enumerate its operands.
if (DILocation *L = I.getDebugLoc())
for (const Metadata *Op : L->operands())
EnumerateMetadata(&F, Op);
}
}
// Optimize constant ordering.
OptimizeConstants(FirstConstant, Values.size());
// Organize metadata ordering.
organizeMetadata();
}
unsigned ValueEnumerator::getInstructionID(const Instruction *Inst) const {
InstructionMapType::const_iterator I = InstructionMap.find(Inst);
assert(I != InstructionMap.end() && "Instruction is not mapped!");
2010-08-25 19:09:50 +02:00
return I->second;
}
unsigned ValueEnumerator::getComdatID(const Comdat *C) const {
unsigned ComdatID = Comdats.idFor(C);
assert(ComdatID && "Comdat not found!");
return ComdatID;
}
void ValueEnumerator::setInstructionID(const Instruction *I) {
InstructionMap[I] = InstructionCount++;
}
unsigned ValueEnumerator::getValueID(const Value *V) const {
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
if (auto *MD = dyn_cast<MetadataAsValue>(V))
return getMetadataID(MD->getMetadata());
ValueMapType::const_iterator I = ValueMap.find(V);
assert(I != ValueMap.end() && "Value not in slotcalculator!");
return I->second-1;
}
LLVM_DUMP_METHOD void ValueEnumerator::dump() const {
print(dbgs(), ValueMap, "Default");
dbgs() << '\n';
print(dbgs(), MetadataMap, "MetaData");
dbgs() << '\n';
}
void ValueEnumerator::print(raw_ostream &OS, const ValueMapType &Map,
const char *Name) const {
OS << "Map Name: " << Name << "\n";
OS << "Size: " << Map.size() << "\n";
for (ValueMapType::const_iterator I = Map.begin(),
E = Map.end(); I != E; ++I) {
const Value *V = I->first;
if (V->hasName())
OS << "Value: " << V->getName();
else
OS << "Value: [null]\n";
V->dump();
OS << " Uses(" << std::distance(V->use_begin(),V->use_end()) << "):";
[C++11] Add range based accessors for the Use-Def chain of a Value. This requires a number of steps. 1) Move value_use_iterator into the Value class as an implementation detail 2) Change it to actually be a *Use* iterator rather than a *User* iterator. 3) Add an adaptor which is a User iterator that always looks through the Use to the User. 4) Wrap these in Value::use_iterator and Value::user_iterator typedefs. 5) Add the range adaptors as Value::uses() and Value::users(). 6) Update *all* of the callers to correctly distinguish between whether they wanted a use_iterator (and to explicitly dig out the User when needed), or a user_iterator which makes the Use itself totally opaque. Because #6 requires churning essentially everything that walked the Use-Def chains, I went ahead and added all of the range adaptors and switched them to range-based loops where appropriate. Also because the renaming requires at least churning every line of code, it didn't make any sense to split these up into multiple commits -- all of which would touch all of the same lies of code. The result is still not quite optimal. The Value::use_iterator is a nice regular iterator, but Value::user_iterator is an iterator over User*s rather than over the User objects themselves. As a consequence, it fits a bit awkwardly into the range-based world and it has the weird extra-dereferencing 'operator->' that so many of our iterators have. I think this could be fixed by providing something which transforms a range of T&s into a range of T*s, but that *can* be separated into another patch, and it isn't yet 100% clear whether this is the right move. However, this change gets us most of the benefit and cleans up a substantial amount of code around Use and User. =] llvm-svn: 203364
2014-03-09 04:16:01 +01:00
for (const Use &U : V->uses()) {
if (&U != &*V->use_begin())
OS << ",";
[C++11] Add range based accessors for the Use-Def chain of a Value. This requires a number of steps. 1) Move value_use_iterator into the Value class as an implementation detail 2) Change it to actually be a *Use* iterator rather than a *User* iterator. 3) Add an adaptor which is a User iterator that always looks through the Use to the User. 4) Wrap these in Value::use_iterator and Value::user_iterator typedefs. 5) Add the range adaptors as Value::uses() and Value::users(). 6) Update *all* of the callers to correctly distinguish between whether they wanted a use_iterator (and to explicitly dig out the User when needed), or a user_iterator which makes the Use itself totally opaque. Because #6 requires churning essentially everything that walked the Use-Def chains, I went ahead and added all of the range adaptors and switched them to range-based loops where appropriate. Also because the renaming requires at least churning every line of code, it didn't make any sense to split these up into multiple commits -- all of which would touch all of the same lies of code. The result is still not quite optimal. The Value::use_iterator is a nice regular iterator, but Value::user_iterator is an iterator over User*s rather than over the User objects themselves. As a consequence, it fits a bit awkwardly into the range-based world and it has the weird extra-dereferencing 'operator->' that so many of our iterators have. I think this could be fixed by providing something which transforms a range of T&s into a range of T*s, but that *can* be separated into another patch, and it isn't yet 100% clear whether this is the right move. However, this change gets us most of the benefit and cleans up a substantial amount of code around Use and User. =] llvm-svn: 203364
2014-03-09 04:16:01 +01:00
if(U->hasName())
OS << " " << U->getName();
else
OS << " [null]";
}
OS << "\n\n";
}
}
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
void ValueEnumerator::print(raw_ostream &OS, const MetadataMapType &Map,
const char *Name) const {
OS << "Map Name: " << Name << "\n";
OS << "Size: " << Map.size() << "\n";
for (auto I = Map.begin(), E = Map.end(); I != E; ++I) {
const Metadata *MD = I->first;
2016-04-02 17:22:57 +02:00
OS << "Metadata: slot = " << I->second.ID << "\n";
OS << "Metadata: function = " << I->second.F << "\n";
MD->print(OS);
2016-04-02 17:22:57 +02:00
OS << "\n";
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
}
}
/// OptimizeConstants - Reorder constant pool for denser encoding.
void ValueEnumerator::OptimizeConstants(unsigned CstStart, unsigned CstEnd) {
if (CstStart == CstEnd || CstStart+1 == CstEnd) return;
if (ShouldPreserveUseListOrder)
// Optimizing constants makes the use-list order difficult to predict.
// Disable it for now when trying to preserve the order.
return;
std::stable_sort(Values.begin() + CstStart, Values.begin() + CstEnd,
[this](const std::pair<const Value *, unsigned> &LHS,
const std::pair<const Value *, unsigned> &RHS) {
// Sort by plane.
if (LHS.first->getType() != RHS.first->getType())
return getTypeID(LHS.first->getType()) < getTypeID(RHS.first->getType());
// Then by frequency.
return LHS.second > RHS.second;
});
// Ensure that integer and vector of integer constants are at the start of the
// constant pool. This is important so that GEP structure indices come before
// gep constant exprs.
std::stable_partition(Values.begin() + CstStart, Values.begin() + CstEnd,
isIntOrIntVectorValue);
// Rebuild the modified portion of ValueMap.
for (; CstStart != CstEnd; ++CstStart)
ValueMap[Values[CstStart].first] = CstStart+1;
}
/// EnumerateValueSymbolTable - Insert all of the values in the specified symbol
/// table into the values table.
void ValueEnumerator::EnumerateValueSymbolTable(const ValueSymbolTable &VST) {
for (ValueSymbolTable::const_iterator VI = VST.begin(), VE = VST.end();
VI != VE; ++VI)
EnumerateValue(VI->getValue());
}
/// Insert all of the values referenced by named metadata in the specified
/// module.
void ValueEnumerator::EnumerateNamedMetadata(const Module &M) {
for (const auto &I : M.named_metadata())
EnumerateNamedMDNode(&I);
}
void ValueEnumerator::EnumerateNamedMDNode(const NamedMDNode *MD) {
for (unsigned i = 0, e = MD->getNumOperands(); i != e; ++i)
2016-04-02 17:22:57 +02:00
EnumerateMetadata(nullptr, MD->getOperand(i));
}
unsigned ValueEnumerator::getMetadataFunctionID(const Function *F) const {
return F ? getValueID(F) + 1 : 0;
}
void ValueEnumerator::EnumerateMetadata(const Function *F, const Metadata *MD) {
EnumerateMetadata(getMetadataFunctionID(F), MD);
}
void ValueEnumerator::EnumerateFunctionLocalMetadata(
const Function &F, const LocalAsMetadata *Local) {
EnumerateFunctionLocalMetadata(getMetadataFunctionID(&F), Local);
}
void ValueEnumerator::dropFunctionFromMetadata(
MetadataMapType::value_type &FirstMD) {
SmallVector<const MDNode *, 64> Worklist;
auto push = [this, &Worklist](MetadataMapType::value_type &MD) {
auto &Entry = MD.second;
// Nothing to do if this metadata isn't tagged.
if (!Entry.F)
return;
// Drop the function tag.
Entry.F = 0;
// If this is has an ID and is an MDNode, then its operands have entries as
// well. We need to drop the function from them too.
if (Entry.ID)
if (auto *N = dyn_cast<MDNode>(MD.first))
Worklist.push_back(N);
};
push(FirstMD);
while (!Worklist.empty())
for (const Metadata *Op : Worklist.pop_back_val()->operands()) {
2016-04-02 17:22:57 +02:00
if (!Op)
continue;
auto MD = MetadataMap.find(Op);
if (MD != MetadataMap.end())
push(*MD);
}
}
2016-04-02 17:22:57 +02:00
void ValueEnumerator::EnumerateMetadata(unsigned F, const Metadata *MD) {
BitcodeWriter: Emit uniqued subgraphs after all distinct nodes Since forward references for uniqued node operands are expensive (and those for distinct node operands are cheap due to DistinctMDOperandPlaceholder), minimize forward references in uniqued node operands. Moreover, guarantee that when a cycle is broken by a distinct node, none of the uniqued nodes have any forward references. In ValueEnumerator::EnumerateMetadata, enumerate uniqued node subgraphs first, delaying distinct nodes until all uniqued nodes have been handled. This guarantees that uniqued nodes only have forward references when there is a uniquing cycle (since r267276 changed ValueEnumerator::organizeMetadata to partition distinct nodes in front of uniqued nodes as a post-pass). Note that a single uniqued subgraph can hit multiple distinct nodes at its leaves. Ideally these would themselves be emitted in post-order, but this commit doesn't attempt that; I think it requires an extra pass through the edges, which I'm not convinced is worth it (since DistinctMDOperandPlaceholder makes forward references quite cheap between distinct nodes). I've added two testcases: - test/Bitcode/mdnodes-distinct-in-post-order.ll is just like test/Bitcode/mdnodes-in-post-order.ll, except with distinct nodes instead of uniqued ones. This confirms that, in the absence of uniqued nodes, distinct nodes are still emitted in post-order. - test/Bitcode/mdnodes-distinct-nodes-break-cycles.ll is the minimal example where a naive post-order traversal would cause one uniqued node to forward-reference another. IOW, it's the motivating test. llvm-svn: 267278
2016-04-23 06:59:22 +02:00
// It's vital for reader efficiency that uniqued subgraphs are done in
// post-order; it's expensive when their operands have forward references.
// If a distinct node is referenced from a uniqued node, it'll be delayed
// until the uniqued subgraph has been completely traversed.
SmallVector<const MDNode *, 32> DelayedDistinctNodes;
// Start by enumerating MD, and then work through its transitive operands in
// post-order. This requires a depth-first search.
ValueMapper/Enumerator: Clean up code in post-order traversals, NFC Re-layer the functions in the new (i.e., newly correct) post-order traversals in ValueEnumerator (r266947) and ValueMapper (r266949). Instead of adding a node to the worklist in a helper function and returning a flag to say what happened, return the node itself. This makes the code way cleaner: the worklist is local to the main function, there is no flag for an early loop exit (since we can cleanly bury the loop), and it's perfectly clear when pointers into the worklist might be invalidated. I'm fixing both algorithms in the same commit to avoid repeating the commit message; if you take the time to understand one the other should be easy. The diff itself isn't entirely obvious since the traversals have some noise (i.e., things to do), but here's the high-level change: auto helper = [&WL](T *Op) { auto helper = [](T **&I, T **E) { => while (I != E) { if (shouldVisit(Op)) { T *Op = *I++; WL.push(Op, Op->begin()); if (shouldVisit(Op)) { return true; return Op; } } return false; return nullptr; }; }; => WL.push(S, S->begin()); WL.push(S, S->begin()); while (!empty()) { while (!empty()) { auto *N = WL.top().N; auto *N = WL.top().N; auto *&I = WL.top().I; auto *&I = WL.top().I; bool DidChange = false; while (I != N->end()) if (helper(*I++)) { => if (T *Op = helper(I, N->end()) { DidChange = true; WL.push(Op, Op->begin()); break; continue; } } if (DidChange) continue; POT.push(WL.pop()); => POT.push(WL.pop()); } } Thanks to Mehdi for helping me find a better way to layer this. llvm-svn: 267099
2016-04-22 04:33:06 +02:00
SmallVector<std::pair<const MDNode *, MDNode::op_iterator>, 32> Worklist;
if (const MDNode *N = enumerateMetadataImpl(F, MD))
Worklist.push_back(std::make_pair(N, N->op_begin()));
while (!Worklist.empty()) {
const MDNode *N = Worklist.back().first;
// Enumerate operands until we hit a new node. We need to traverse these
// nodes' operands before visiting the rest of N's operands.
MDNode::op_iterator I = std::find_if(
Worklist.back().second, N->op_end(),
[&](const Metadata *MD) { return enumerateMetadataImpl(F, MD); });
if (I != N->op_end()) {
auto *Op = cast<MDNode>(*I);
Worklist.back().second = ++I;
BitcodeWriter: Emit uniqued subgraphs after all distinct nodes Since forward references for uniqued node operands are expensive (and those for distinct node operands are cheap due to DistinctMDOperandPlaceholder), minimize forward references in uniqued node operands. Moreover, guarantee that when a cycle is broken by a distinct node, none of the uniqued nodes have any forward references. In ValueEnumerator::EnumerateMetadata, enumerate uniqued node subgraphs first, delaying distinct nodes until all uniqued nodes have been handled. This guarantees that uniqued nodes only have forward references when there is a uniquing cycle (since r267276 changed ValueEnumerator::organizeMetadata to partition distinct nodes in front of uniqued nodes as a post-pass). Note that a single uniqued subgraph can hit multiple distinct nodes at its leaves. Ideally these would themselves be emitted in post-order, but this commit doesn't attempt that; I think it requires an extra pass through the edges, which I'm not convinced is worth it (since DistinctMDOperandPlaceholder makes forward references quite cheap between distinct nodes). I've added two testcases: - test/Bitcode/mdnodes-distinct-in-post-order.ll is just like test/Bitcode/mdnodes-in-post-order.ll, except with distinct nodes instead of uniqued ones. This confirms that, in the absence of uniqued nodes, distinct nodes are still emitted in post-order. - test/Bitcode/mdnodes-distinct-nodes-break-cycles.ll is the minimal example where a naive post-order traversal would cause one uniqued node to forward-reference another. IOW, it's the motivating test. llvm-svn: 267278
2016-04-23 06:59:22 +02:00
// Delay traversing Op if it's a distinct node and N is uniqued.
if (Op->isDistinct() && !N->isDistinct())
DelayedDistinctNodes.push_back(Op);
else
Worklist.push_back(std::make_pair(Op, Op->op_begin()));
continue;
ValueMapper/Enumerator: Clean up code in post-order traversals, NFC Re-layer the functions in the new (i.e., newly correct) post-order traversals in ValueEnumerator (r266947) and ValueMapper (r266949). Instead of adding a node to the worklist in a helper function and returning a flag to say what happened, return the node itself. This makes the code way cleaner: the worklist is local to the main function, there is no flag for an early loop exit (since we can cleanly bury the loop), and it's perfectly clear when pointers into the worklist might be invalidated. I'm fixing both algorithms in the same commit to avoid repeating the commit message; if you take the time to understand one the other should be easy. The diff itself isn't entirely obvious since the traversals have some noise (i.e., things to do), but here's the high-level change: auto helper = [&WL](T *Op) { auto helper = [](T **&I, T **E) { => while (I != E) { if (shouldVisit(Op)) { T *Op = *I++; WL.push(Op, Op->begin()); if (shouldVisit(Op)) { return true; return Op; } } return false; return nullptr; }; }; => WL.push(S, S->begin()); WL.push(S, S->begin()); while (!empty()) { while (!empty()) { auto *N = WL.top().N; auto *N = WL.top().N; auto *&I = WL.top().I; auto *&I = WL.top().I; bool DidChange = false; while (I != N->end()) if (helper(*I++)) { => if (T *Op = helper(I, N->end()) { DidChange = true; WL.push(Op, Op->begin()); break; continue; } } if (DidChange) continue; POT.push(WL.pop()); => POT.push(WL.pop()); } } Thanks to Mehdi for helping me find a better way to layer this. llvm-svn: 267099
2016-04-22 04:33:06 +02:00
}
// All the operands have been visited. Now assign an ID.
Worklist.pop_back();
MDs.push_back(N);
MetadataMap[N].ID = MDs.size();
BitcodeWriter: Emit uniqued subgraphs after all distinct nodes Since forward references for uniqued node operands are expensive (and those for distinct node operands are cheap due to DistinctMDOperandPlaceholder), minimize forward references in uniqued node operands. Moreover, guarantee that when a cycle is broken by a distinct node, none of the uniqued nodes have any forward references. In ValueEnumerator::EnumerateMetadata, enumerate uniqued node subgraphs first, delaying distinct nodes until all uniqued nodes have been handled. This guarantees that uniqued nodes only have forward references when there is a uniquing cycle (since r267276 changed ValueEnumerator::organizeMetadata to partition distinct nodes in front of uniqued nodes as a post-pass). Note that a single uniqued subgraph can hit multiple distinct nodes at its leaves. Ideally these would themselves be emitted in post-order, but this commit doesn't attempt that; I think it requires an extra pass through the edges, which I'm not convinced is worth it (since DistinctMDOperandPlaceholder makes forward references quite cheap between distinct nodes). I've added two testcases: - test/Bitcode/mdnodes-distinct-in-post-order.ll is just like test/Bitcode/mdnodes-in-post-order.ll, except with distinct nodes instead of uniqued ones. This confirms that, in the absence of uniqued nodes, distinct nodes are still emitted in post-order. - test/Bitcode/mdnodes-distinct-nodes-break-cycles.ll is the minimal example where a naive post-order traversal would cause one uniqued node to forward-reference another. IOW, it's the motivating test. llvm-svn: 267278
2016-04-23 06:59:22 +02:00
// Flush out any delayed distinct nodes; these are all the distinct nodes
// that are leaves in last uniqued subgraph.
if (Worklist.empty() || Worklist.back().first->isDistinct()) {
for (const MDNode *N : DelayedDistinctNodes)
Worklist.push_back(std::make_pair(N, N->op_begin()));
DelayedDistinctNodes.clear();
}
}
}
ValueMapper/Enumerator: Clean up code in post-order traversals, NFC Re-layer the functions in the new (i.e., newly correct) post-order traversals in ValueEnumerator (r266947) and ValueMapper (r266949). Instead of adding a node to the worklist in a helper function and returning a flag to say what happened, return the node itself. This makes the code way cleaner: the worklist is local to the main function, there is no flag for an early loop exit (since we can cleanly bury the loop), and it's perfectly clear when pointers into the worklist might be invalidated. I'm fixing both algorithms in the same commit to avoid repeating the commit message; if you take the time to understand one the other should be easy. The diff itself isn't entirely obvious since the traversals have some noise (i.e., things to do), but here's the high-level change: auto helper = [&WL](T *Op) { auto helper = [](T **&I, T **E) { => while (I != E) { if (shouldVisit(Op)) { T *Op = *I++; WL.push(Op, Op->begin()); if (shouldVisit(Op)) { return true; return Op; } } return false; return nullptr; }; }; => WL.push(S, S->begin()); WL.push(S, S->begin()); while (!empty()) { while (!empty()) { auto *N = WL.top().N; auto *N = WL.top().N; auto *&I = WL.top().I; auto *&I = WL.top().I; bool DidChange = false; while (I != N->end()) if (helper(*I++)) { => if (T *Op = helper(I, N->end()) { DidChange = true; WL.push(Op, Op->begin()); break; continue; } } if (DidChange) continue; POT.push(WL.pop()); => POT.push(WL.pop()); } } Thanks to Mehdi for helping me find a better way to layer this. llvm-svn: 267099
2016-04-22 04:33:06 +02:00
const MDNode *ValueEnumerator::enumerateMetadataImpl(unsigned F, const Metadata *MD) {
if (!MD)
ValueMapper/Enumerator: Clean up code in post-order traversals, NFC Re-layer the functions in the new (i.e., newly correct) post-order traversals in ValueEnumerator (r266947) and ValueMapper (r266949). Instead of adding a node to the worklist in a helper function and returning a flag to say what happened, return the node itself. This makes the code way cleaner: the worklist is local to the main function, there is no flag for an early loop exit (since we can cleanly bury the loop), and it's perfectly clear when pointers into the worklist might be invalidated. I'm fixing both algorithms in the same commit to avoid repeating the commit message; if you take the time to understand one the other should be easy. The diff itself isn't entirely obvious since the traversals have some noise (i.e., things to do), but here's the high-level change: auto helper = [&WL](T *Op) { auto helper = [](T **&I, T **E) { => while (I != E) { if (shouldVisit(Op)) { T *Op = *I++; WL.push(Op, Op->begin()); if (shouldVisit(Op)) { return true; return Op; } } return false; return nullptr; }; }; => WL.push(S, S->begin()); WL.push(S, S->begin()); while (!empty()) { while (!empty()) { auto *N = WL.top().N; auto *N = WL.top().N; auto *&I = WL.top().I; auto *&I = WL.top().I; bool DidChange = false; while (I != N->end()) if (helper(*I++)) { => if (T *Op = helper(I, N->end()) { DidChange = true; WL.push(Op, Op->begin()); break; continue; } } if (DidChange) continue; POT.push(WL.pop()); => POT.push(WL.pop()); } } Thanks to Mehdi for helping me find a better way to layer this. llvm-svn: 267099
2016-04-22 04:33:06 +02:00
return nullptr;
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
assert(
(isa<MDNode>(MD) || isa<MDString>(MD) || isa<ConstantAsMetadata>(MD)) &&
"Invalid metadata kind");
auto Insertion = MetadataMap.insert(std::make_pair(MD, MDIndex(F)));
MDIndex &Entry = Insertion.first->second;
if (!Insertion.second) {
// Already mapped. If F doesn't match the function tag, drop it.
if (Entry.hasDifferentFunction(F))
dropFunctionFromMetadata(*Insertion.first);
ValueMapper/Enumerator: Clean up code in post-order traversals, NFC Re-layer the functions in the new (i.e., newly correct) post-order traversals in ValueEnumerator (r266947) and ValueMapper (r266949). Instead of adding a node to the worklist in a helper function and returning a flag to say what happened, return the node itself. This makes the code way cleaner: the worklist is local to the main function, there is no flag for an early loop exit (since we can cleanly bury the loop), and it's perfectly clear when pointers into the worklist might be invalidated. I'm fixing both algorithms in the same commit to avoid repeating the commit message; if you take the time to understand one the other should be easy. The diff itself isn't entirely obvious since the traversals have some noise (i.e., things to do), but here's the high-level change: auto helper = [&WL](T *Op) { auto helper = [](T **&I, T **E) { => while (I != E) { if (shouldVisit(Op)) { T *Op = *I++; WL.push(Op, Op->begin()); if (shouldVisit(Op)) { return true; return Op; } } return false; return nullptr; }; }; => WL.push(S, S->begin()); WL.push(S, S->begin()); while (!empty()) { while (!empty()) { auto *N = WL.top().N; auto *N = WL.top().N; auto *&I = WL.top().I; auto *&I = WL.top().I; bool DidChange = false; while (I != N->end()) if (helper(*I++)) { => if (T *Op = helper(I, N->end()) { DidChange = true; WL.push(Op, Op->begin()); break; continue; } } if (DidChange) continue; POT.push(WL.pop()); => POT.push(WL.pop()); } } Thanks to Mehdi for helping me find a better way to layer this. llvm-svn: 267099
2016-04-22 04:33:06 +02:00
return nullptr;
}
ValueMapper/Enumerator: Clean up code in post-order traversals, NFC Re-layer the functions in the new (i.e., newly correct) post-order traversals in ValueEnumerator (r266947) and ValueMapper (r266949). Instead of adding a node to the worklist in a helper function and returning a flag to say what happened, return the node itself. This makes the code way cleaner: the worklist is local to the main function, there is no flag for an early loop exit (since we can cleanly bury the loop), and it's perfectly clear when pointers into the worklist might be invalidated. I'm fixing both algorithms in the same commit to avoid repeating the commit message; if you take the time to understand one the other should be easy. The diff itself isn't entirely obvious since the traversals have some noise (i.e., things to do), but here's the high-level change: auto helper = [&WL](T *Op) { auto helper = [](T **&I, T **E) { => while (I != E) { if (shouldVisit(Op)) { T *Op = *I++; WL.push(Op, Op->begin()); if (shouldVisit(Op)) { return true; return Op; } } return false; return nullptr; }; }; => WL.push(S, S->begin()); WL.push(S, S->begin()); while (!empty()) { while (!empty()) { auto *N = WL.top().N; auto *N = WL.top().N; auto *&I = WL.top().I; auto *&I = WL.top().I; bool DidChange = false; while (I != N->end()) if (helper(*I++)) { => if (T *Op = helper(I, N->end()) { DidChange = true; WL.push(Op, Op->begin()); break; continue; } } if (DidChange) continue; POT.push(WL.pop()); => POT.push(WL.pop()); } } Thanks to Mehdi for helping me find a better way to layer this. llvm-svn: 267099
2016-04-22 04:33:06 +02:00
// Don't assign IDs to metadata nodes.
if (auto *N = dyn_cast<MDNode>(MD))
return N;
2016-04-02 17:22:57 +02:00
// Save the metadata.
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
MDs.push_back(MD);
Entry.ID = MDs.size();
// Enumerate the constant, if any.
if (auto *C = dyn_cast<ConstantAsMetadata>(MD))
EnumerateValue(C->getValue());
ValueMapper/Enumerator: Clean up code in post-order traversals, NFC Re-layer the functions in the new (i.e., newly correct) post-order traversals in ValueEnumerator (r266947) and ValueMapper (r266949). Instead of adding a node to the worklist in a helper function and returning a flag to say what happened, return the node itself. This makes the code way cleaner: the worklist is local to the main function, there is no flag for an early loop exit (since we can cleanly bury the loop), and it's perfectly clear when pointers into the worklist might be invalidated. I'm fixing both algorithms in the same commit to avoid repeating the commit message; if you take the time to understand one the other should be easy. The diff itself isn't entirely obvious since the traversals have some noise (i.e., things to do), but here's the high-level change: auto helper = [&WL](T *Op) { auto helper = [](T **&I, T **E) { => while (I != E) { if (shouldVisit(Op)) { T *Op = *I++; WL.push(Op, Op->begin()); if (shouldVisit(Op)) { return true; return Op; } } return false; return nullptr; }; }; => WL.push(S, S->begin()); WL.push(S, S->begin()); while (!empty()) { while (!empty()) { auto *N = WL.top().N; auto *N = WL.top().N; auto *&I = WL.top().I; auto *&I = WL.top().I; bool DidChange = false; while (I != N->end()) if (helper(*I++)) { => if (T *Op = helper(I, N->end()) { DidChange = true; WL.push(Op, Op->begin()); break; continue; } } if (DidChange) continue; POT.push(WL.pop()); => POT.push(WL.pop()); } } Thanks to Mehdi for helping me find a better way to layer this. llvm-svn: 267099
2016-04-22 04:33:06 +02:00
return nullptr;
}
/// EnumerateFunctionLocalMetadataa - Incorporate function-local metadata
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
/// information reachable from the metadata.
void ValueEnumerator::EnumerateFunctionLocalMetadata(
2016-04-02 17:22:57 +02:00
unsigned F, const LocalAsMetadata *Local) {
assert(F && "Expected a function");
// Check to see if it's already in!
2016-04-02 17:22:57 +02:00
MDIndex &Index = MetadataMap[Local];
if (Index.ID) {
assert(Index.F == F && "Expected the same function");
return;
2016-04-02 17:22:57 +02:00
}
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
MDs.push_back(Local);
2016-04-02 17:22:57 +02:00
Index.F = F;
Index.ID = MDs.size();
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
EnumerateValue(Local->getValue());
}
static unsigned getMetadataTypeOrder(const Metadata *MD) {
// Strings are emitted in bulk and must come first.
if (isa<MDString>(MD))
return 0;
// ConstantAsMetadata doesn't reference anything. We may as well shuffle it
// to the front since we can detect it.
auto *N = dyn_cast<MDNode>(MD);
if (!N)
return 1;
// The reader is fast forward references for distinct node operands, but slow
// when uniqued operands are unresolved.
return N->isDistinct() ? 2 : 3;
}
void ValueEnumerator::organizeMetadata() {
2016-04-02 17:22:57 +02:00
assert(MetadataMap.size() == MDs.size() &&
"Metadata map and vector out of sync");
if (MDs.empty())
return;
// Copy out the index information from MetadataMap in order to choose a new
// order.
SmallVector<MDIndex, 64> Order;
Order.reserve(MetadataMap.size());
for (const Metadata *MD : MDs)
Order.push_back(MetadataMap.lookup(MD));
// Partition:
// - by function, then
// - by isa<MDString>
// and then sort by the original/current ID. Since the IDs are guaranteed to
// be unique, the result of std::sort will be deterministic. There's no need
// for std::stable_sort.
std::sort(Order.begin(), Order.end(), [this](MDIndex LHS, MDIndex RHS) {
return std::make_tuple(LHS.F, getMetadataTypeOrder(LHS.get(MDs)), LHS.ID) <
std::make_tuple(RHS.F, getMetadataTypeOrder(RHS.get(MDs)), RHS.ID);
2016-04-02 17:22:57 +02:00
});
// Rebuild MDs, index the metadata ranges for each function in FunctionMDs,
// and fix up MetadataMap.
std::vector<const Metadata *> OldMDs = std::move(MDs);
MDs.reserve(OldMDs.size());
for (unsigned I = 0, E = Order.size(); I != E && !Order[I].F; ++I) {
auto *MD = Order[I].get(OldMDs);
MDs.push_back(MD);
MetadataMap[MD].ID = I + 1;
if (isa<MDString>(MD))
++NumMDStrings;
}
// Return early if there's nothing for the functions.
if (MDs.size() == Order.size())
return;
2016-04-02 17:22:57 +02:00
// Build the function metadata ranges.
MDRange R;
FunctionMDs.reserve(OldMDs.size());
unsigned PrevF = 0;
for (unsigned I = MDs.size(), E = Order.size(), ID = MDs.size(); I != E;
++I) {
unsigned F = Order[I].F;
if (!PrevF) {
PrevF = F;
} else if (PrevF != F) {
R.Last = FunctionMDs.size();
std::swap(R, FunctionMDInfo[PrevF]);
R.First = FunctionMDs.size();
ID = MDs.size();
PrevF = F;
}
2016-04-02 17:22:57 +02:00
auto *MD = Order[I].get(OldMDs);
FunctionMDs.push_back(MD);
MetadataMap[MD].ID = ++ID;
if (isa<MDString>(MD))
++R.NumStrings;
}
R.Last = FunctionMDs.size();
FunctionMDInfo[PrevF] = R;
}
void ValueEnumerator::incorporateFunctionMetadata(const Function &F) {
NumModuleMDs = MDs.size();
auto R = FunctionMDInfo.lookup(getValueID(&F) + 1);
NumMDStrings = R.NumStrings;
MDs.insert(MDs.end(), FunctionMDs.begin() + R.First,
FunctionMDs.begin() + R.Last);
}
void ValueEnumerator::EnumerateValue(const Value *V) {
2009-12-31 01:51:46 +01:00
assert(!V->getType()->isVoidTy() && "Can't insert void values!");
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
assert(!isa<MetadataAsValue>(V) && "EnumerateValue doesn't handle Metadata!");
// Check to see if it's already in!
unsigned &ValueID = ValueMap[V];
if (ValueID) {
// Increment use count.
Values[ValueID-1].second++;
return;
}
if (auto *GO = dyn_cast<GlobalObject>(V))
if (const Comdat *C = GO->getComdat())
Comdats.insert(C);
// Enumerate the type of this value.
EnumerateType(V->getType());
if (const Constant *C = dyn_cast<Constant>(V)) {
if (isa<GlobalValue>(C)) {
// Initializers for globals are handled explicitly elsewhere.
} else if (C->getNumOperands()) {
// If a constant has operands, enumerate them. This makes sure that if a
// constant has uses (for example an array of const ints), that they are
// inserted also.
// We prefer to enumerate them with values before we enumerate the user
// itself. This makes it more likely that we can avoid forward references
// in the reader. We know that there can be no cycles in the constants
// graph that don't go through a global variable.
for (User::const_op_iterator I = C->op_begin(), E = C->op_end();
I != E; ++I)
if (!isa<BasicBlock>(*I)) // Don't enumerate BB operand to BlockAddress.
EnumerateValue(*I);
// Finally, add the value. Doing this could make the ValueID reference be
// dangling, don't reuse it.
Values.push_back(std::make_pair(V, 1U));
ValueMap[V] = Values.size();
return;
}
}
// Add the value.
Values.push_back(std::make_pair(V, 1U));
ValueID = Values.size();
}
void ValueEnumerator::EnumerateType(Type *Ty) {
unsigned *TypeID = &TypeMap[Ty];
// We've already seen this type.
if (*TypeID)
return;
// If it is a non-anonymous struct, mark the type as being visited so that we
// don't recursively visit it. This is safe because we allow forward
// references of these in the bitcode reader.
if (StructType *STy = dyn_cast<StructType>(Ty))
if (!STy->isLiteral())
*TypeID = ~0U;
// Enumerate all of the subtypes before we enumerate this type. This ensures
// that the type will be enumerated in an order that can be directly built.
for (Type *SubTy : Ty->subtypes())
EnumerateType(SubTy);
// Refresh the TypeID pointer in case the table rehashed.
TypeID = &TypeMap[Ty];
// Check to see if we got the pointer another way. This can happen when
// enumerating recursive types that hit the base case deeper than they start.
//
// If this is actually a struct that we are treating as forward ref'able,
// then emit the definition now that all of its contents are available.
if (*TypeID && *TypeID != ~0U)
return;
// Add this type now that its contents are all happily enumerated.
Types.push_back(Ty);
*TypeID = Types.size();
}
// Enumerate the types for the specified value. If the value is a constant,
// walk through it, enumerating the types of the constant.
void ValueEnumerator::EnumerateOperandType(const Value *V) {
EnumerateType(V->getType());
assert(!isa<MetadataAsValue>(V) && "Unexpected metadata operand");
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
const Constant *C = dyn_cast<Constant>(V);
if (!C)
return;
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
// If this constant is already enumerated, ignore it, we know its type must
// be enumerated.
if (ValueMap.count(C))
return;
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
// This constant may have operands, make sure to enumerate the types in
// them.
for (const Value *Op : C->operands()) {
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
// Don't enumerate basic blocks here, this happens as operands to
// blockaddress.
if (isa<BasicBlock>(Op))
continue;
EnumerateOperandType(Op);
}
}
void ValueEnumerator::EnumerateAttributes(AttributeSet PAL) {
if (PAL.isEmpty()) return; // null is always 0.
// Do a lookup.
unsigned &Entry = AttributeMap[PAL];
if (Entry == 0) {
// Never saw this before, add it.
Attribute.push_back(PAL);
Entry = Attribute.size();
}
// Do lookups for all attribute groups.
for (unsigned i = 0, e = PAL.getNumSlots(); i != e; ++i) {
AttributeSet AS = PAL.getSlotAttributes(i);
unsigned &Entry = AttributeGroupMap[AS];
if (Entry == 0) {
AttributeGroups.push_back(AS);
Entry = AttributeGroups.size();
}
}
}
void ValueEnumerator::incorporateFunction(const Function &F) {
InstructionCount = 0;
NumModuleValues = Values.size();
2016-04-02 17:22:57 +02:00
// Add global metadata to the function block. This doesn't include
// LocalAsMetadata.
incorporateFunctionMetadata(F);
// Adding function arguments to the value table.
for (const auto &I : F.args())
EnumerateValue(&I);
FirstFuncConstantID = Values.size();
// Add all function-level constants to the value table.
for (const BasicBlock &BB : F) {
for (const Instruction &I : BB)
for (const Use &OI : I.operands()) {
if ((isa<Constant>(OI) && !isa<GlobalValue>(OI)) || isa<InlineAsm>(OI))
EnumerateValue(OI);
}
BasicBlocks.push_back(&BB);
ValueMap[&BB] = BasicBlocks.size();
}
// Optimize the constant layout.
OptimizeConstants(FirstFuncConstantID, Values.size());
// Add the function's parameter attributes so they are available for use in
// the function's instruction.
EnumerateAttributes(F.getAttributes());
FirstInstID = Values.size();
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
SmallVector<LocalAsMetadata *, 8> FnLocalMDVector;
// Add all of the instructions.
for (const BasicBlock &BB : F) {
for (const Instruction &I : BB) {
for (const Use &OI : I.operands()) {
if (auto *MD = dyn_cast<MetadataAsValue>(&OI))
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
if (auto *Local = dyn_cast<LocalAsMetadata>(MD->getMetadata()))
// Enumerate metadata after the instructions they might refer to.
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
FnLocalMDVector.push_back(Local);
}
if (!I.getType()->isVoidTy())
EnumerateValue(&I);
}
}
// Add all of the function-local metadata.
for (unsigned i = 0, e = FnLocalMDVector.size(); i != e; ++i)
2016-04-02 17:22:57 +02:00
EnumerateFunctionLocalMetadata(F, FnLocalMDVector[i]);
}
void ValueEnumerator::purgeFunction() {
/// Remove purged values from the ValueMap.
for (unsigned i = NumModuleValues, e = Values.size(); i != e; ++i)
ValueMap.erase(Values[i].first);
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
for (unsigned i = NumModuleMDs, e = MDs.size(); i != e; ++i)
MetadataMap.erase(MDs[i]);
for (unsigned i = 0, e = BasicBlocks.size(); i != e; ++i)
ValueMap.erase(BasicBlocks[i]);
Values.resize(NumModuleValues);
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
MDs.resize(NumModuleMDs);
BasicBlocks.clear();
2016-04-02 17:22:57 +02:00
NumMDStrings = 0;
}
static void IncorporateFunctionInfoGlobalBBIDs(const Function *F,
DenseMap<const BasicBlock*, unsigned> &IDMap) {
unsigned Counter = 0;
for (const BasicBlock &BB : *F)
IDMap[&BB] = ++Counter;
}
/// getGlobalBasicBlockID - This returns the function-specific ID for the
/// specified basic block. This is relatively expensive information, so it
/// should only be used by rare constructs such as address-of-label.
unsigned ValueEnumerator::getGlobalBasicBlockID(const BasicBlock *BB) const {
unsigned &Idx = GlobalBasicBlockIDs[BB];
if (Idx != 0)
return Idx-1;
IncorporateFunctionInfoGlobalBBIDs(BB->getParent(), GlobalBasicBlockIDs);
return getGlobalBasicBlockID(BB);
}
uint64_t ValueEnumerator::computeBitsRequiredForTypeIndicies() const {
return Log2_32_Ceil(getTypes().size() + 1);
}