1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-24 21:42:54 +02:00
llvm-mirror/lib/Transforms/Utils/ValueMapper.cpp

494 lines
19 KiB
C++
Raw Normal View History

//===- ValueMapper.cpp - Interface shared by lib/Transforms/Utils ---------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file defines the MapValue function, which is shared by various parts of
// the lib/Transforms/Utils library.
//
//===----------------------------------------------------------------------===//
#include "llvm/Transforms/Utils/ValueMapper.h"
#include "llvm/IR/CallSite.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/InlineAsm.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/Metadata.h"
#include "llvm/IR/Operator.h"
using namespace llvm;
// Out of line method to get vtable etc for class.
void ValueMapTypeRemapper::anchor() {}
void ValueMaterializer::anchor() {}
void ValueMaterializer::materializeInitFor(GlobalValue *New, GlobalValue *Old) {
}
Value *llvm::MapValue(const Value *V, ValueToValueMapTy &VM, RemapFlags Flags,
ValueMapTypeRemapper *TypeMapper,
ValueMaterializer *Materializer) {
ValueToValueMapTy::iterator I = VM.find(V);
// If the value already exists in the map, use it.
if (I != VM.end() && I->second) return I->second;
// If we have a materializer and it can materialize a value, use that.
if (Materializer) {
if (Value *NewV =
Materializer->materializeDeclFor(const_cast<Value *>(V))) {
VM[V] = NewV;
if (auto *NewGV = dyn_cast<GlobalValue>(NewV))
Materializer->materializeInitFor(
NewGV, const_cast<GlobalValue *>(cast<GlobalValue>(V)));
return NewV;
}
}
// Global values do not need to be seeded into the VM if they
// are using the identity mapping.
Fix mapping of unmaterialized global values during metadata linking Summary: The patch to move metadata linking after global value linking didn't correctly map unmaterialized global values to null as desired. They were in fact mapped to the source copy. It largely worked by accident since most module linker clients destroyed the source module which caused the source GVs to be replaced by null, but caused a failure with LTO linking on Windows: http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20151109/312869.html The problem is that a null return value from materializeValueFor is handled by mapping the value to self. This is the desired behavior when materializeValueFor is passed a non-GlobalValue. The problem is how to distinguish that case from the case where we really do want to map to null. This patch addresses this by passing in a new flag to the value mapper indicating that unmapped global values should be mapped to null. Other Value types are handled as before. Note that the documented behavior of asserting on unmapped values when the flag RF_IgnoreMissingValues isn't set is currently disabled with FIXME notes due to bootstrap failures. I modified these disabled asserts so when they are eventually enabled again it won't assert for the unmapped values when the new RF_NullMapMissingGlobalValues flag is set. I also considered using a callback into the value materializer, but a flag seemed cleaner given that there are already existing flags. I also considered modifying materializeValueFor to return the input value when we want to map to source and then treat a null return to mean map to null. However, there are other value materializer subclasses that implement materializeValueFor, and they would all need to be audited and the return values possibly changed, which seemed error-prone. Reviewers: dexonsmith, joker.eph Subscribers: pcc, llvm-commits Differential Revision: http://reviews.llvm.org/D14682 llvm-svn: 253170
2015-11-15 15:50:14 +01:00
if (isa<GlobalValue>(V)) {
if (Flags & RF_NullMapMissingGlobalValues) {
assert(!(Flags & RF_IgnoreMissingEntries) &&
"Illegal to specify both RF_NullMapMissingGlobalValues and "
"RF_IgnoreMissingEntries");
return nullptr;
}
return VM[V] = const_cast<Value*>(V);
Fix mapping of unmaterialized global values during metadata linking Summary: The patch to move metadata linking after global value linking didn't correctly map unmaterialized global values to null as desired. They were in fact mapped to the source copy. It largely worked by accident since most module linker clients destroyed the source module which caused the source GVs to be replaced by null, but caused a failure with LTO linking on Windows: http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20151109/312869.html The problem is that a null return value from materializeValueFor is handled by mapping the value to self. This is the desired behavior when materializeValueFor is passed a non-GlobalValue. The problem is how to distinguish that case from the case where we really do want to map to null. This patch addresses this by passing in a new flag to the value mapper indicating that unmapped global values should be mapped to null. Other Value types are handled as before. Note that the documented behavior of asserting on unmapped values when the flag RF_IgnoreMissingValues isn't set is currently disabled with FIXME notes due to bootstrap failures. I modified these disabled asserts so when they are eventually enabled again it won't assert for the unmapped values when the new RF_NullMapMissingGlobalValues flag is set. I also considered using a callback into the value materializer, but a flag seemed cleaner given that there are already existing flags. I also considered modifying materializeValueFor to return the input value when we want to map to source and then treat a null return to mean map to null. However, there are other value materializer subclasses that implement materializeValueFor, and they would all need to be audited and the return values possibly changed, which seemed error-prone. Reviewers: dexonsmith, joker.eph Subscribers: pcc, llvm-commits Differential Revision: http://reviews.llvm.org/D14682 llvm-svn: 253170
2015-11-15 15:50:14 +01:00
}
if (const InlineAsm *IA = dyn_cast<InlineAsm>(V)) {
// Inline asm may need *type* remapping.
FunctionType *NewTy = IA->getFunctionType();
if (TypeMapper) {
NewTy = cast<FunctionType>(TypeMapper->remapType(NewTy));
if (NewTy != IA->getFunctionType())
V = InlineAsm::get(NewTy, IA->getAsmString(), IA->getConstraintString(),
IA->hasSideEffects(), IA->isAlignStack());
}
return VM[V] = const_cast<Value*>(V);
}
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 (const auto *MDV = dyn_cast<MetadataAsValue>(V)) {
const Metadata *MD = MDV->getMetadata();
// If this is a module-level metadata and we know that nothing at the module
// level is changing, then use an identity mapping.
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 (!isa<LocalAsMetadata>(MD) && (Flags & RF_NoModuleLevelChanges))
return VM[V] = const_cast<Value *>(V);
auto *MappedMD = MapMetadata(MD, VM, Flags, TypeMapper, Materializer);
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 (MD == MappedMD || (!MappedMD && (Flags & RF_IgnoreMissingEntries)))
return VM[V] = const_cast<Value *>(V);
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
// FIXME: This assert crashes during bootstrap, but I think it should be
// correct. For now, just match behaviour from before the metadata/value
// split.
//
Fix mapping of unmaterialized global values during metadata linking Summary: The patch to move metadata linking after global value linking didn't correctly map unmaterialized global values to null as desired. They were in fact mapped to the source copy. It largely worked by accident since most module linker clients destroyed the source module which caused the source GVs to be replaced by null, but caused a failure with LTO linking on Windows: http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20151109/312869.html The problem is that a null return value from materializeValueFor is handled by mapping the value to self. This is the desired behavior when materializeValueFor is passed a non-GlobalValue. The problem is how to distinguish that case from the case where we really do want to map to null. This patch addresses this by passing in a new flag to the value mapper indicating that unmapped global values should be mapped to null. Other Value types are handled as before. Note that the documented behavior of asserting on unmapped values when the flag RF_IgnoreMissingValues isn't set is currently disabled with FIXME notes due to bootstrap failures. I modified these disabled asserts so when they are eventually enabled again it won't assert for the unmapped values when the new RF_NullMapMissingGlobalValues flag is set. I also considered using a callback into the value materializer, but a flag seemed cleaner given that there are already existing flags. I also considered modifying materializeValueFor to return the input value when we want to map to source and then treat a null return to mean map to null. However, there are other value materializer subclasses that implement materializeValueFor, and they would all need to be audited and the return values possibly changed, which seemed error-prone. Reviewers: dexonsmith, joker.eph Subscribers: pcc, llvm-commits Differential Revision: http://reviews.llvm.org/D14682 llvm-svn: 253170
2015-11-15 15:50:14 +01:00
// assert((MappedMD || (Flags & RF_NullMapMissingGlobalValues)) &&
// "Referenced metadata value not in value map");
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
return VM[V] = MetadataAsValue::get(V->getContext(), MappedMD);
}
// Okay, this either must be a constant (which may or may not be mappable) or
// is something that is not in the mapping table.
Constant *C = const_cast<Constant*>(dyn_cast<Constant>(V));
if (!C)
return nullptr;
if (BlockAddress *BA = dyn_cast<BlockAddress>(C)) {
Function *F =
cast<Function>(MapValue(BA->getFunction(), VM, Flags, TypeMapper, Materializer));
BasicBlock *BB = cast_or_null<BasicBlock>(MapValue(BA->getBasicBlock(), VM,
Flags, TypeMapper, Materializer));
return VM[V] = BlockAddress::get(F, BB ? BB : BA->getBasicBlock());
}
// Otherwise, we have some other constant to remap. Start by checking to see
// if all operands have an identity remapping.
unsigned OpNo = 0, NumOperands = C->getNumOperands();
Value *Mapped = nullptr;
for (; OpNo != NumOperands; ++OpNo) {
Value *Op = C->getOperand(OpNo);
Mapped = MapValue(Op, VM, Flags, TypeMapper, Materializer);
if (Mapped != C) break;
}
// See if the type mapper wants to remap the type as well.
Type *NewTy = C->getType();
if (TypeMapper)
NewTy = TypeMapper->remapType(NewTy);
// If the result type and all operands match up, then just insert an identity
// mapping.
if (OpNo == NumOperands && NewTy == C->getType())
return VM[V] = C;
// Okay, we need to create a new constant. We've already processed some or
// all of the operands, set them all up now.
SmallVector<Constant*, 8> Ops;
Ops.reserve(NumOperands);
for (unsigned j = 0; j != OpNo; ++j)
Ops.push_back(cast<Constant>(C->getOperand(j)));
// If one of the operands mismatch, push it and the other mapped operands.
if (OpNo != NumOperands) {
Ops.push_back(cast<Constant>(Mapped));
// Map the rest of the operands that aren't processed yet.
for (++OpNo; OpNo != NumOperands; ++OpNo)
Ops.push_back(MapValue(cast<Constant>(C->getOperand(OpNo)), VM,
Flags, TypeMapper, Materializer));
}
Type *NewSrcTy = nullptr;
if (TypeMapper)
if (auto *GEPO = dyn_cast<GEPOperator>(C))
NewSrcTy = TypeMapper->remapType(GEPO->getSourceElementType());
if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C))
return VM[V] = CE->getWithOperands(Ops, NewTy, false, NewSrcTy);
if (isa<ConstantArray>(C))
return VM[V] = ConstantArray::get(cast<ArrayType>(NewTy), Ops);
if (isa<ConstantStruct>(C))
return VM[V] = ConstantStruct::get(cast<StructType>(NewTy), Ops);
if (isa<ConstantVector>(C))
return VM[V] = ConstantVector::get(Ops);
// If this is a no-operand constant, it must be because the type was remapped.
if (isa<UndefValue>(C))
return VM[V] = UndefValue::get(NewTy);
if (isa<ConstantAggregateZero>(C))
return VM[V] = ConstantAggregateZero::get(NewTy);
assert(isa<ConstantPointerNull>(C));
return VM[V] = ConstantPointerNull::get(cast<PointerType>(NewTy));
}
static Metadata *mapToMetadata(ValueToValueMapTy &VM, const Metadata *Key,
Metadata *Val, ValueMaterializer *Materializer,
RemapFlags Flags) {
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
VM.MD()[Key].reset(Val);
if (Materializer && !(Flags & RF_HaveUnmaterializedMetadata)) {
auto *N = dyn_cast_or_null<MDNode>(Val);
// Need to invoke this once we have non-temporary MD.
if (!N || !N->isTemporary())
Materializer->replaceTemporaryMetadata(Key, Val);
}
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
return Val;
}
static Metadata *mapToSelf(ValueToValueMapTy &VM, const Metadata *MD,
ValueMaterializer *Materializer, RemapFlags Flags) {
return mapToMetadata(VM, MD, const_cast<Metadata *>(MD), Materializer, Flags);
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
}
static Metadata *MapMetadataImpl(const Metadata *MD,
ValueMapper: Rotate distinct node remapping algorithm Rotate the algorithm for remapping distinct nodes in order to simplify how uniquing cycles get resolved. This removes some of the recursion, and, most importantly, exposes all uniquing cycles at the top-level. Besides being a little more efficient -- temporary MDNodes won't live as long -- the clearer logic should help protect against bugs like those fixed in r243961 and r243976. What are uniquing cycles? Why do they present challenges when remapping metadata? !0 = !{!1} !1 = !{!0} !0 and !1 form a simple uniquing cycle. When remapping from one metadata graph to another, every uniquing cycle gets "duplicated" through a dance: !0-temp = !{!1?} ; map(!0): clone !0, VM[!0] = !0-temp !1-temp = !{!0?} ; ..map(!1): clone !1, VM[!1] = !1-temp !1-temp = !{!0-temp} ; ..map(!1): remap !1's operands !2 = !{!0-temp} ; ..map(!1): uniquify: !1-temp => !2 !0-temp = !{!2} ; map(!0): remap !0's operands !3 = !{!2} ; map(!0): uniquify: !0-temp => !3 ; Result !2 = !{!3} !3 = !{!2} (In the two "uniquify" steps above, the operands of !X-temp are compared to the operands of !X. If they're the same, then !X-temp gets RAUW'ed to !X; if they're different, then !X-temp is promoted to a new unique node. The latter case always hits in for uniquing cycles, so we duplicate all the nodes involved.) Why is this a problem? Uniquable Metadata nodes that have temporary node as transitive operands keep RAUW support until the temporary nodes get finalized. With non-cycles, this happens automatically: when a uniquable node's count of unresolved operands drops to zero, it immediately sheds its own RAUW support (possibly triggering the same in any node that references it). However, uniquing cycles create a reference cycle, and uniqued nodes that transitively reference a uniquing cycle are "stuck" in an unresolved state until someone calls `MDNode::resolveCycles()` on a node in the unresolved subgraph. Distinct nodes should help here (and mostly do): since they aren't uniqued anywhere, they are guaranteed not to be RAUW'ed. They effectively form a barrier between uniqued nodes, breaking some uniquing cycles, and shielding uniqued nodes from uniquing cycles. Unfortunately, with this barrier in place, the unresolved subgraph(s) can be disjoint from the top-level node. The mapping algorithm needs to find at least one representative from each disjoint subgraph. But which nodes are *stuck*, and which will get resolved automatically? And which nodes are in the unresolved subgraph? The old logic was conservative. This commit rotates the logic for distinct nodes, so that we have access to unresolved nodes at the top-level call to `llvm::MapMetadata()`. Each time we return to the top-level, we know that all temporaries have been RAUW'ed away. Here, it's safe (and necessary) to call `resolveCycles()` immediately on unresolved operands. This should also perform better than the old algorithm. The recursion stack is shorter, temporary nodes don't live as long, and there are fewer tracking references to unresolved nodes. As the debug info graph introduces more 'distinct' nodes, remapping should incrementally get cheaper and cheaper. Aside from possible performance improvements (and reduced cruft in the `LLVMContext`), there should be no functionality change here. llvm-svn: 244181
2015-08-06 01:52:42 +02:00
SmallVectorImpl<MDNode *> &DistinctWorklist,
ValueToValueMapTy &VM, RemapFlags Flags,
ValueMapTypeRemapper *TypeMapper,
ValueMaterializer *Materializer);
static Metadata *mapMetadataOp(Metadata *Op,
ValueMapper: Rotate distinct node remapping algorithm Rotate the algorithm for remapping distinct nodes in order to simplify how uniquing cycles get resolved. This removes some of the recursion, and, most importantly, exposes all uniquing cycles at the top-level. Besides being a little more efficient -- temporary MDNodes won't live as long -- the clearer logic should help protect against bugs like those fixed in r243961 and r243976. What are uniquing cycles? Why do they present challenges when remapping metadata? !0 = !{!1} !1 = !{!0} !0 and !1 form a simple uniquing cycle. When remapping from one metadata graph to another, every uniquing cycle gets "duplicated" through a dance: !0-temp = !{!1?} ; map(!0): clone !0, VM[!0] = !0-temp !1-temp = !{!0?} ; ..map(!1): clone !1, VM[!1] = !1-temp !1-temp = !{!0-temp} ; ..map(!1): remap !1's operands !2 = !{!0-temp} ; ..map(!1): uniquify: !1-temp => !2 !0-temp = !{!2} ; map(!0): remap !0's operands !3 = !{!2} ; map(!0): uniquify: !0-temp => !3 ; Result !2 = !{!3} !3 = !{!2} (In the two "uniquify" steps above, the operands of !X-temp are compared to the operands of !X. If they're the same, then !X-temp gets RAUW'ed to !X; if they're different, then !X-temp is promoted to a new unique node. The latter case always hits in for uniquing cycles, so we duplicate all the nodes involved.) Why is this a problem? Uniquable Metadata nodes that have temporary node as transitive operands keep RAUW support until the temporary nodes get finalized. With non-cycles, this happens automatically: when a uniquable node's count of unresolved operands drops to zero, it immediately sheds its own RAUW support (possibly triggering the same in any node that references it). However, uniquing cycles create a reference cycle, and uniqued nodes that transitively reference a uniquing cycle are "stuck" in an unresolved state until someone calls `MDNode::resolveCycles()` on a node in the unresolved subgraph. Distinct nodes should help here (and mostly do): since they aren't uniqued anywhere, they are guaranteed not to be RAUW'ed. They effectively form a barrier between uniqued nodes, breaking some uniquing cycles, and shielding uniqued nodes from uniquing cycles. Unfortunately, with this barrier in place, the unresolved subgraph(s) can be disjoint from the top-level node. The mapping algorithm needs to find at least one representative from each disjoint subgraph. But which nodes are *stuck*, and which will get resolved automatically? And which nodes are in the unresolved subgraph? The old logic was conservative. This commit rotates the logic for distinct nodes, so that we have access to unresolved nodes at the top-level call to `llvm::MapMetadata()`. Each time we return to the top-level, we know that all temporaries have been RAUW'ed away. Here, it's safe (and necessary) to call `resolveCycles()` immediately on unresolved operands. This should also perform better than the old algorithm. The recursion stack is shorter, temporary nodes don't live as long, and there are fewer tracking references to unresolved nodes. As the debug info graph introduces more 'distinct' nodes, remapping should incrementally get cheaper and cheaper. Aside from possible performance improvements (and reduced cruft in the `LLVMContext`), there should be no functionality change here. llvm-svn: 244181
2015-08-06 01:52:42 +02:00
SmallVectorImpl<MDNode *> &DistinctWorklist,
ValueToValueMapTy &VM, RemapFlags Flags,
ValueMapTypeRemapper *TypeMapper,
ValueMaterializer *Materializer) {
if (!Op)
return nullptr;
if (Materializer && !Materializer->isMetadataNeeded(Op))
return nullptr;
ValueMapper: Rotate distinct node remapping algorithm Rotate the algorithm for remapping distinct nodes in order to simplify how uniquing cycles get resolved. This removes some of the recursion, and, most importantly, exposes all uniquing cycles at the top-level. Besides being a little more efficient -- temporary MDNodes won't live as long -- the clearer logic should help protect against bugs like those fixed in r243961 and r243976. What are uniquing cycles? Why do they present challenges when remapping metadata? !0 = !{!1} !1 = !{!0} !0 and !1 form a simple uniquing cycle. When remapping from one metadata graph to another, every uniquing cycle gets "duplicated" through a dance: !0-temp = !{!1?} ; map(!0): clone !0, VM[!0] = !0-temp !1-temp = !{!0?} ; ..map(!1): clone !1, VM[!1] = !1-temp !1-temp = !{!0-temp} ; ..map(!1): remap !1's operands !2 = !{!0-temp} ; ..map(!1): uniquify: !1-temp => !2 !0-temp = !{!2} ; map(!0): remap !0's operands !3 = !{!2} ; map(!0): uniquify: !0-temp => !3 ; Result !2 = !{!3} !3 = !{!2} (In the two "uniquify" steps above, the operands of !X-temp are compared to the operands of !X. If they're the same, then !X-temp gets RAUW'ed to !X; if they're different, then !X-temp is promoted to a new unique node. The latter case always hits in for uniquing cycles, so we duplicate all the nodes involved.) Why is this a problem? Uniquable Metadata nodes that have temporary node as transitive operands keep RAUW support until the temporary nodes get finalized. With non-cycles, this happens automatically: when a uniquable node's count of unresolved operands drops to zero, it immediately sheds its own RAUW support (possibly triggering the same in any node that references it). However, uniquing cycles create a reference cycle, and uniqued nodes that transitively reference a uniquing cycle are "stuck" in an unresolved state until someone calls `MDNode::resolveCycles()` on a node in the unresolved subgraph. Distinct nodes should help here (and mostly do): since they aren't uniqued anywhere, they are guaranteed not to be RAUW'ed. They effectively form a barrier between uniqued nodes, breaking some uniquing cycles, and shielding uniqued nodes from uniquing cycles. Unfortunately, with this barrier in place, the unresolved subgraph(s) can be disjoint from the top-level node. The mapping algorithm needs to find at least one representative from each disjoint subgraph. But which nodes are *stuck*, and which will get resolved automatically? And which nodes are in the unresolved subgraph? The old logic was conservative. This commit rotates the logic for distinct nodes, so that we have access to unresolved nodes at the top-level call to `llvm::MapMetadata()`. Each time we return to the top-level, we know that all temporaries have been RAUW'ed away. Here, it's safe (and necessary) to call `resolveCycles()` immediately on unresolved operands. This should also perform better than the old algorithm. The recursion stack is shorter, temporary nodes don't live as long, and there are fewer tracking references to unresolved nodes. As the debug info graph introduces more 'distinct' nodes, remapping should incrementally get cheaper and cheaper. Aside from possible performance improvements (and reduced cruft in the `LLVMContext`), there should be no functionality change here. llvm-svn: 244181
2015-08-06 01:52:42 +02:00
if (Metadata *MappedOp = MapMetadataImpl(Op, DistinctWorklist, VM, Flags,
TypeMapper, Materializer))
return MappedOp;
// Use identity map if MappedOp is null and we can ignore missing entries.
if (Flags & RF_IgnoreMissingEntries)
return Op;
// FIXME: This assert crashes during bootstrap, but I think it should be
// correct. For now, just match behaviour from before the metadata/value
// split.
//
Fix mapping of unmaterialized global values during metadata linking Summary: The patch to move metadata linking after global value linking didn't correctly map unmaterialized global values to null as desired. They were in fact mapped to the source copy. It largely worked by accident since most module linker clients destroyed the source module which caused the source GVs to be replaced by null, but caused a failure with LTO linking on Windows: http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20151109/312869.html The problem is that a null return value from materializeValueFor is handled by mapping the value to self. This is the desired behavior when materializeValueFor is passed a non-GlobalValue. The problem is how to distinguish that case from the case where we really do want to map to null. This patch addresses this by passing in a new flag to the value mapper indicating that unmapped global values should be mapped to null. Other Value types are handled as before. Note that the documented behavior of asserting on unmapped values when the flag RF_IgnoreMissingValues isn't set is currently disabled with FIXME notes due to bootstrap failures. I modified these disabled asserts so when they are eventually enabled again it won't assert for the unmapped values when the new RF_NullMapMissingGlobalValues flag is set. I also considered using a callback into the value materializer, but a flag seemed cleaner given that there are already existing flags. I also considered modifying materializeValueFor to return the input value when we want to map to source and then treat a null return to mean map to null. However, there are other value materializer subclasses that implement materializeValueFor, and they would all need to be audited and the return values possibly changed, which seemed error-prone. Reviewers: dexonsmith, joker.eph Subscribers: pcc, llvm-commits Differential Revision: http://reviews.llvm.org/D14682 llvm-svn: 253170
2015-11-15 15:50:14 +01:00
// assert((Flags & RF_NullMapMissingGlobalValues) &&
// "Referenced metadata not in value map!");
return nullptr;
}
/// Resolve uniquing cycles involving the given metadata.
static void resolveCycles(Metadata *MD, bool AllowTemps) {
if (auto *N = dyn_cast_or_null<MDNode>(MD)) {
if (AllowTemps && N->isTemporary())
return;
if (!N->isResolved()) {
if (AllowTemps)
// Note that this will drop RAUW support on any temporaries, which
// blocks uniquing. If this ends up being an issue, in the future
// we can experiment with delaying resolving these nodes until
// after metadata is fully materialized (i.e. when linking metadata
// as a postpass after function importing).
N->resolveNonTemporaries();
else
N->resolveCycles();
}
}
}
/// Remap the operands of an MDNode.
///
/// If \c Node is temporary, uniquing cycles are ignored. If \c Node is
/// distinct, uniquing cycles are resolved as they're found.
///
/// \pre \c Node.isDistinct() or \c Node.isTemporary().
static bool remapOperands(MDNode &Node,
ValueMapper: Rotate distinct node remapping algorithm Rotate the algorithm for remapping distinct nodes in order to simplify how uniquing cycles get resolved. This removes some of the recursion, and, most importantly, exposes all uniquing cycles at the top-level. Besides being a little more efficient -- temporary MDNodes won't live as long -- the clearer logic should help protect against bugs like those fixed in r243961 and r243976. What are uniquing cycles? Why do they present challenges when remapping metadata? !0 = !{!1} !1 = !{!0} !0 and !1 form a simple uniquing cycle. When remapping from one metadata graph to another, every uniquing cycle gets "duplicated" through a dance: !0-temp = !{!1?} ; map(!0): clone !0, VM[!0] = !0-temp !1-temp = !{!0?} ; ..map(!1): clone !1, VM[!1] = !1-temp !1-temp = !{!0-temp} ; ..map(!1): remap !1's operands !2 = !{!0-temp} ; ..map(!1): uniquify: !1-temp => !2 !0-temp = !{!2} ; map(!0): remap !0's operands !3 = !{!2} ; map(!0): uniquify: !0-temp => !3 ; Result !2 = !{!3} !3 = !{!2} (In the two "uniquify" steps above, the operands of !X-temp are compared to the operands of !X. If they're the same, then !X-temp gets RAUW'ed to !X; if they're different, then !X-temp is promoted to a new unique node. The latter case always hits in for uniquing cycles, so we duplicate all the nodes involved.) Why is this a problem? Uniquable Metadata nodes that have temporary node as transitive operands keep RAUW support until the temporary nodes get finalized. With non-cycles, this happens automatically: when a uniquable node's count of unresolved operands drops to zero, it immediately sheds its own RAUW support (possibly triggering the same in any node that references it). However, uniquing cycles create a reference cycle, and uniqued nodes that transitively reference a uniquing cycle are "stuck" in an unresolved state until someone calls `MDNode::resolveCycles()` on a node in the unresolved subgraph. Distinct nodes should help here (and mostly do): since they aren't uniqued anywhere, they are guaranteed not to be RAUW'ed. They effectively form a barrier between uniqued nodes, breaking some uniquing cycles, and shielding uniqued nodes from uniquing cycles. Unfortunately, with this barrier in place, the unresolved subgraph(s) can be disjoint from the top-level node. The mapping algorithm needs to find at least one representative from each disjoint subgraph. But which nodes are *stuck*, and which will get resolved automatically? And which nodes are in the unresolved subgraph? The old logic was conservative. This commit rotates the logic for distinct nodes, so that we have access to unresolved nodes at the top-level call to `llvm::MapMetadata()`. Each time we return to the top-level, we know that all temporaries have been RAUW'ed away. Here, it's safe (and necessary) to call `resolveCycles()` immediately on unresolved operands. This should also perform better than the old algorithm. The recursion stack is shorter, temporary nodes don't live as long, and there are fewer tracking references to unresolved nodes. As the debug info graph introduces more 'distinct' nodes, remapping should incrementally get cheaper and cheaper. Aside from possible performance improvements (and reduced cruft in the `LLVMContext`), there should be no functionality change here. llvm-svn: 244181
2015-08-06 01:52:42 +02:00
SmallVectorImpl<MDNode *> &DistinctWorklist,
ValueToValueMapTy &VM, RemapFlags Flags,
ValueMapTypeRemapper *TypeMapper,
ValueMaterializer *Materializer) {
assert(!Node.isUniqued() && "Expected temporary or distinct node");
const bool IsDistinct = Node.isDistinct();
bool AnyChanged = false;
for (unsigned I = 0, E = Node.getNumOperands(); I != E; ++I) {
Metadata *Old = Node.getOperand(I);
ValueMapper: Rotate distinct node remapping algorithm Rotate the algorithm for remapping distinct nodes in order to simplify how uniquing cycles get resolved. This removes some of the recursion, and, most importantly, exposes all uniquing cycles at the top-level. Besides being a little more efficient -- temporary MDNodes won't live as long -- the clearer logic should help protect against bugs like those fixed in r243961 and r243976. What are uniquing cycles? Why do they present challenges when remapping metadata? !0 = !{!1} !1 = !{!0} !0 and !1 form a simple uniquing cycle. When remapping from one metadata graph to another, every uniquing cycle gets "duplicated" through a dance: !0-temp = !{!1?} ; map(!0): clone !0, VM[!0] = !0-temp !1-temp = !{!0?} ; ..map(!1): clone !1, VM[!1] = !1-temp !1-temp = !{!0-temp} ; ..map(!1): remap !1's operands !2 = !{!0-temp} ; ..map(!1): uniquify: !1-temp => !2 !0-temp = !{!2} ; map(!0): remap !0's operands !3 = !{!2} ; map(!0): uniquify: !0-temp => !3 ; Result !2 = !{!3} !3 = !{!2} (In the two "uniquify" steps above, the operands of !X-temp are compared to the operands of !X. If they're the same, then !X-temp gets RAUW'ed to !X; if they're different, then !X-temp is promoted to a new unique node. The latter case always hits in for uniquing cycles, so we duplicate all the nodes involved.) Why is this a problem? Uniquable Metadata nodes that have temporary node as transitive operands keep RAUW support until the temporary nodes get finalized. With non-cycles, this happens automatically: when a uniquable node's count of unresolved operands drops to zero, it immediately sheds its own RAUW support (possibly triggering the same in any node that references it). However, uniquing cycles create a reference cycle, and uniqued nodes that transitively reference a uniquing cycle are "stuck" in an unresolved state until someone calls `MDNode::resolveCycles()` on a node in the unresolved subgraph. Distinct nodes should help here (and mostly do): since they aren't uniqued anywhere, they are guaranteed not to be RAUW'ed. They effectively form a barrier between uniqued nodes, breaking some uniquing cycles, and shielding uniqued nodes from uniquing cycles. Unfortunately, with this barrier in place, the unresolved subgraph(s) can be disjoint from the top-level node. The mapping algorithm needs to find at least one representative from each disjoint subgraph. But which nodes are *stuck*, and which will get resolved automatically? And which nodes are in the unresolved subgraph? The old logic was conservative. This commit rotates the logic for distinct nodes, so that we have access to unresolved nodes at the top-level call to `llvm::MapMetadata()`. Each time we return to the top-level, we know that all temporaries have been RAUW'ed away. Here, it's safe (and necessary) to call `resolveCycles()` immediately on unresolved operands. This should also perform better than the old algorithm. The recursion stack is shorter, temporary nodes don't live as long, and there are fewer tracking references to unresolved nodes. As the debug info graph introduces more 'distinct' nodes, remapping should incrementally get cheaper and cheaper. Aside from possible performance improvements (and reduced cruft in the `LLVMContext`), there should be no functionality change here. llvm-svn: 244181
2015-08-06 01:52:42 +02:00
Metadata *New = mapMetadataOp(Old, DistinctWorklist, VM, Flags, TypeMapper,
Materializer);
if (Old != New) {
AnyChanged = true;
Node.replaceOperandWith(I, New);
// Resolve uniquing cycles underneath distinct nodes on the fly so they
// don't infect later operands.
if (IsDistinct)
resolveCycles(New, Flags & RF_HaveUnmaterializedMetadata);
}
}
return AnyChanged;
}
/// Map a distinct MDNode.
///
/// Whether distinct nodes change is independent of their operands. If \a
/// RF_MoveDistinctMDs, then they are reused, and their operands remapped in
/// place; effectively, they're moved from one graph to another. Otherwise,
/// they're cloned/duplicated, and the new copy's operands are remapped.
static Metadata *mapDistinctNode(const MDNode *Node,
ValueMapper: Rotate distinct node remapping algorithm Rotate the algorithm for remapping distinct nodes in order to simplify how uniquing cycles get resolved. This removes some of the recursion, and, most importantly, exposes all uniquing cycles at the top-level. Besides being a little more efficient -- temporary MDNodes won't live as long -- the clearer logic should help protect against bugs like those fixed in r243961 and r243976. What are uniquing cycles? Why do they present challenges when remapping metadata? !0 = !{!1} !1 = !{!0} !0 and !1 form a simple uniquing cycle. When remapping from one metadata graph to another, every uniquing cycle gets "duplicated" through a dance: !0-temp = !{!1?} ; map(!0): clone !0, VM[!0] = !0-temp !1-temp = !{!0?} ; ..map(!1): clone !1, VM[!1] = !1-temp !1-temp = !{!0-temp} ; ..map(!1): remap !1's operands !2 = !{!0-temp} ; ..map(!1): uniquify: !1-temp => !2 !0-temp = !{!2} ; map(!0): remap !0's operands !3 = !{!2} ; map(!0): uniquify: !0-temp => !3 ; Result !2 = !{!3} !3 = !{!2} (In the two "uniquify" steps above, the operands of !X-temp are compared to the operands of !X. If they're the same, then !X-temp gets RAUW'ed to !X; if they're different, then !X-temp is promoted to a new unique node. The latter case always hits in for uniquing cycles, so we duplicate all the nodes involved.) Why is this a problem? Uniquable Metadata nodes that have temporary node as transitive operands keep RAUW support until the temporary nodes get finalized. With non-cycles, this happens automatically: when a uniquable node's count of unresolved operands drops to zero, it immediately sheds its own RAUW support (possibly triggering the same in any node that references it). However, uniquing cycles create a reference cycle, and uniqued nodes that transitively reference a uniquing cycle are "stuck" in an unresolved state until someone calls `MDNode::resolveCycles()` on a node in the unresolved subgraph. Distinct nodes should help here (and mostly do): since they aren't uniqued anywhere, they are guaranteed not to be RAUW'ed. They effectively form a barrier between uniqued nodes, breaking some uniquing cycles, and shielding uniqued nodes from uniquing cycles. Unfortunately, with this barrier in place, the unresolved subgraph(s) can be disjoint from the top-level node. The mapping algorithm needs to find at least one representative from each disjoint subgraph. But which nodes are *stuck*, and which will get resolved automatically? And which nodes are in the unresolved subgraph? The old logic was conservative. This commit rotates the logic for distinct nodes, so that we have access to unresolved nodes at the top-level call to `llvm::MapMetadata()`. Each time we return to the top-level, we know that all temporaries have been RAUW'ed away. Here, it's safe (and necessary) to call `resolveCycles()` immediately on unresolved operands. This should also perform better than the old algorithm. The recursion stack is shorter, temporary nodes don't live as long, and there are fewer tracking references to unresolved nodes. As the debug info graph introduces more 'distinct' nodes, remapping should incrementally get cheaper and cheaper. Aside from possible performance improvements (and reduced cruft in the `LLVMContext`), there should be no functionality change here. llvm-svn: 244181
2015-08-06 01:52:42 +02:00
SmallVectorImpl<MDNode *> &DistinctWorklist,
ValueToValueMapTy &VM, RemapFlags Flags,
ValueMapTypeRemapper *TypeMapper,
ValueMaterializer *Materializer) {
assert(Node->isDistinct() && "Expected distinct node");
MDNode *NewMD;
if (Flags & RF_MoveDistinctMDs)
NewMD = const_cast<MDNode *>(Node);
else
NewMD = MDNode::replaceWithDistinct(Node->clone());
ValueMapper: Rotate distinct node remapping algorithm Rotate the algorithm for remapping distinct nodes in order to simplify how uniquing cycles get resolved. This removes some of the recursion, and, most importantly, exposes all uniquing cycles at the top-level. Besides being a little more efficient -- temporary MDNodes won't live as long -- the clearer logic should help protect against bugs like those fixed in r243961 and r243976. What are uniquing cycles? Why do they present challenges when remapping metadata? !0 = !{!1} !1 = !{!0} !0 and !1 form a simple uniquing cycle. When remapping from one metadata graph to another, every uniquing cycle gets "duplicated" through a dance: !0-temp = !{!1?} ; map(!0): clone !0, VM[!0] = !0-temp !1-temp = !{!0?} ; ..map(!1): clone !1, VM[!1] = !1-temp !1-temp = !{!0-temp} ; ..map(!1): remap !1's operands !2 = !{!0-temp} ; ..map(!1): uniquify: !1-temp => !2 !0-temp = !{!2} ; map(!0): remap !0's operands !3 = !{!2} ; map(!0): uniquify: !0-temp => !3 ; Result !2 = !{!3} !3 = !{!2} (In the two "uniquify" steps above, the operands of !X-temp are compared to the operands of !X. If they're the same, then !X-temp gets RAUW'ed to !X; if they're different, then !X-temp is promoted to a new unique node. The latter case always hits in for uniquing cycles, so we duplicate all the nodes involved.) Why is this a problem? Uniquable Metadata nodes that have temporary node as transitive operands keep RAUW support until the temporary nodes get finalized. With non-cycles, this happens automatically: when a uniquable node's count of unresolved operands drops to zero, it immediately sheds its own RAUW support (possibly triggering the same in any node that references it). However, uniquing cycles create a reference cycle, and uniqued nodes that transitively reference a uniquing cycle are "stuck" in an unresolved state until someone calls `MDNode::resolveCycles()` on a node in the unresolved subgraph. Distinct nodes should help here (and mostly do): since they aren't uniqued anywhere, they are guaranteed not to be RAUW'ed. They effectively form a barrier between uniqued nodes, breaking some uniquing cycles, and shielding uniqued nodes from uniquing cycles. Unfortunately, with this barrier in place, the unresolved subgraph(s) can be disjoint from the top-level node. The mapping algorithm needs to find at least one representative from each disjoint subgraph. But which nodes are *stuck*, and which will get resolved automatically? And which nodes are in the unresolved subgraph? The old logic was conservative. This commit rotates the logic for distinct nodes, so that we have access to unresolved nodes at the top-level call to `llvm::MapMetadata()`. Each time we return to the top-level, we know that all temporaries have been RAUW'ed away. Here, it's safe (and necessary) to call `resolveCycles()` immediately on unresolved operands. This should also perform better than the old algorithm. The recursion stack is shorter, temporary nodes don't live as long, and there are fewer tracking references to unresolved nodes. As the debug info graph introduces more 'distinct' nodes, remapping should incrementally get cheaper and cheaper. Aside from possible performance improvements (and reduced cruft in the `LLVMContext`), there should be no functionality change here. llvm-svn: 244181
2015-08-06 01:52:42 +02:00
// Remap operands later.
DistinctWorklist.push_back(NewMD);
return mapToMetadata(VM, Node, NewMD, Materializer, Flags);
}
/// \brief Map a uniqued MDNode.
///
/// Uniqued nodes may not need to be recreated (they may map to themselves).
static Metadata *mapUniquedNode(const MDNode *Node,
ValueMapper: Rotate distinct node remapping algorithm Rotate the algorithm for remapping distinct nodes in order to simplify how uniquing cycles get resolved. This removes some of the recursion, and, most importantly, exposes all uniquing cycles at the top-level. Besides being a little more efficient -- temporary MDNodes won't live as long -- the clearer logic should help protect against bugs like those fixed in r243961 and r243976. What are uniquing cycles? Why do they present challenges when remapping metadata? !0 = !{!1} !1 = !{!0} !0 and !1 form a simple uniquing cycle. When remapping from one metadata graph to another, every uniquing cycle gets "duplicated" through a dance: !0-temp = !{!1?} ; map(!0): clone !0, VM[!0] = !0-temp !1-temp = !{!0?} ; ..map(!1): clone !1, VM[!1] = !1-temp !1-temp = !{!0-temp} ; ..map(!1): remap !1's operands !2 = !{!0-temp} ; ..map(!1): uniquify: !1-temp => !2 !0-temp = !{!2} ; map(!0): remap !0's operands !3 = !{!2} ; map(!0): uniquify: !0-temp => !3 ; Result !2 = !{!3} !3 = !{!2} (In the two "uniquify" steps above, the operands of !X-temp are compared to the operands of !X. If they're the same, then !X-temp gets RAUW'ed to !X; if they're different, then !X-temp is promoted to a new unique node. The latter case always hits in for uniquing cycles, so we duplicate all the nodes involved.) Why is this a problem? Uniquable Metadata nodes that have temporary node as transitive operands keep RAUW support until the temporary nodes get finalized. With non-cycles, this happens automatically: when a uniquable node's count of unresolved operands drops to zero, it immediately sheds its own RAUW support (possibly triggering the same in any node that references it). However, uniquing cycles create a reference cycle, and uniqued nodes that transitively reference a uniquing cycle are "stuck" in an unresolved state until someone calls `MDNode::resolveCycles()` on a node in the unresolved subgraph. Distinct nodes should help here (and mostly do): since they aren't uniqued anywhere, they are guaranteed not to be RAUW'ed. They effectively form a barrier between uniqued nodes, breaking some uniquing cycles, and shielding uniqued nodes from uniquing cycles. Unfortunately, with this barrier in place, the unresolved subgraph(s) can be disjoint from the top-level node. The mapping algorithm needs to find at least one representative from each disjoint subgraph. But which nodes are *stuck*, and which will get resolved automatically? And which nodes are in the unresolved subgraph? The old logic was conservative. This commit rotates the logic for distinct nodes, so that we have access to unresolved nodes at the top-level call to `llvm::MapMetadata()`. Each time we return to the top-level, we know that all temporaries have been RAUW'ed away. Here, it's safe (and necessary) to call `resolveCycles()` immediately on unresolved operands. This should also perform better than the old algorithm. The recursion stack is shorter, temporary nodes don't live as long, and there are fewer tracking references to unresolved nodes. As the debug info graph introduces more 'distinct' nodes, remapping should incrementally get cheaper and cheaper. Aside from possible performance improvements (and reduced cruft in the `LLVMContext`), there should be no functionality change here. llvm-svn: 244181
2015-08-06 01:52:42 +02:00
SmallVectorImpl<MDNode *> &DistinctWorklist,
ValueToValueMapTy &VM, RemapFlags Flags,
2015-01-19 23:40:25 +01:00
ValueMapTypeRemapper *TypeMapper,
ValueMaterializer *Materializer) {
assert(((Flags & RF_HaveUnmaterializedMetadata) || Node->isUniqued()) &&
"Expected uniqued node");
// Create a temporary node and map it upfront in case we have a uniquing
// cycle. If necessary, this mapping will get updated by RAUW logic before
// returning.
auto ClonedMD = Node->clone();
mapToMetadata(VM, Node, ClonedMD.get(), Materializer, Flags);
ValueMapper: Rotate distinct node remapping algorithm Rotate the algorithm for remapping distinct nodes in order to simplify how uniquing cycles get resolved. This removes some of the recursion, and, most importantly, exposes all uniquing cycles at the top-level. Besides being a little more efficient -- temporary MDNodes won't live as long -- the clearer logic should help protect against bugs like those fixed in r243961 and r243976. What are uniquing cycles? Why do they present challenges when remapping metadata? !0 = !{!1} !1 = !{!0} !0 and !1 form a simple uniquing cycle. When remapping from one metadata graph to another, every uniquing cycle gets "duplicated" through a dance: !0-temp = !{!1?} ; map(!0): clone !0, VM[!0] = !0-temp !1-temp = !{!0?} ; ..map(!1): clone !1, VM[!1] = !1-temp !1-temp = !{!0-temp} ; ..map(!1): remap !1's operands !2 = !{!0-temp} ; ..map(!1): uniquify: !1-temp => !2 !0-temp = !{!2} ; map(!0): remap !0's operands !3 = !{!2} ; map(!0): uniquify: !0-temp => !3 ; Result !2 = !{!3} !3 = !{!2} (In the two "uniquify" steps above, the operands of !X-temp are compared to the operands of !X. If they're the same, then !X-temp gets RAUW'ed to !X; if they're different, then !X-temp is promoted to a new unique node. The latter case always hits in for uniquing cycles, so we duplicate all the nodes involved.) Why is this a problem? Uniquable Metadata nodes that have temporary node as transitive operands keep RAUW support until the temporary nodes get finalized. With non-cycles, this happens automatically: when a uniquable node's count of unresolved operands drops to zero, it immediately sheds its own RAUW support (possibly triggering the same in any node that references it). However, uniquing cycles create a reference cycle, and uniqued nodes that transitively reference a uniquing cycle are "stuck" in an unresolved state until someone calls `MDNode::resolveCycles()` on a node in the unresolved subgraph. Distinct nodes should help here (and mostly do): since they aren't uniqued anywhere, they are guaranteed not to be RAUW'ed. They effectively form a barrier between uniqued nodes, breaking some uniquing cycles, and shielding uniqued nodes from uniquing cycles. Unfortunately, with this barrier in place, the unresolved subgraph(s) can be disjoint from the top-level node. The mapping algorithm needs to find at least one representative from each disjoint subgraph. But which nodes are *stuck*, and which will get resolved automatically? And which nodes are in the unresolved subgraph? The old logic was conservative. This commit rotates the logic for distinct nodes, so that we have access to unresolved nodes at the top-level call to `llvm::MapMetadata()`. Each time we return to the top-level, we know that all temporaries have been RAUW'ed away. Here, it's safe (and necessary) to call `resolveCycles()` immediately on unresolved operands. This should also perform better than the old algorithm. The recursion stack is shorter, temporary nodes don't live as long, and there are fewer tracking references to unresolved nodes. As the debug info graph introduces more 'distinct' nodes, remapping should incrementally get cheaper and cheaper. Aside from possible performance improvements (and reduced cruft in the `LLVMContext`), there should be no functionality change here. llvm-svn: 244181
2015-08-06 01:52:42 +02:00
if (!remapOperands(*ClonedMD, DistinctWorklist, VM, Flags, TypeMapper,
Materializer)) {
// No operands changed, so use the original.
ClonedMD->replaceAllUsesWith(const_cast<MDNode *>(Node));
// Even though replaceAllUsesWith would have replaced the value map
// entry, we need to explictly map with the final non-temporary node
// to replace any temporary metadata via the callback.
return mapToSelf(VM, Node, Materializer, Flags);
}
// Uniquify the cloned node. Explicitly map it with the final non-temporary
// node so that replacement of temporary metadata via the callback occurs.
return mapToMetadata(VM, Node,
MDNode::replaceWithUniqued(std::move(ClonedMD)),
Materializer, Flags);
}
static Metadata *MapMetadataImpl(const Metadata *MD,
ValueMapper: Rotate distinct node remapping algorithm Rotate the algorithm for remapping distinct nodes in order to simplify how uniquing cycles get resolved. This removes some of the recursion, and, most importantly, exposes all uniquing cycles at the top-level. Besides being a little more efficient -- temporary MDNodes won't live as long -- the clearer logic should help protect against bugs like those fixed in r243961 and r243976. What are uniquing cycles? Why do they present challenges when remapping metadata? !0 = !{!1} !1 = !{!0} !0 and !1 form a simple uniquing cycle. When remapping from one metadata graph to another, every uniquing cycle gets "duplicated" through a dance: !0-temp = !{!1?} ; map(!0): clone !0, VM[!0] = !0-temp !1-temp = !{!0?} ; ..map(!1): clone !1, VM[!1] = !1-temp !1-temp = !{!0-temp} ; ..map(!1): remap !1's operands !2 = !{!0-temp} ; ..map(!1): uniquify: !1-temp => !2 !0-temp = !{!2} ; map(!0): remap !0's operands !3 = !{!2} ; map(!0): uniquify: !0-temp => !3 ; Result !2 = !{!3} !3 = !{!2} (In the two "uniquify" steps above, the operands of !X-temp are compared to the operands of !X. If they're the same, then !X-temp gets RAUW'ed to !X; if they're different, then !X-temp is promoted to a new unique node. The latter case always hits in for uniquing cycles, so we duplicate all the nodes involved.) Why is this a problem? Uniquable Metadata nodes that have temporary node as transitive operands keep RAUW support until the temporary nodes get finalized. With non-cycles, this happens automatically: when a uniquable node's count of unresolved operands drops to zero, it immediately sheds its own RAUW support (possibly triggering the same in any node that references it). However, uniquing cycles create a reference cycle, and uniqued nodes that transitively reference a uniquing cycle are "stuck" in an unresolved state until someone calls `MDNode::resolveCycles()` on a node in the unresolved subgraph. Distinct nodes should help here (and mostly do): since they aren't uniqued anywhere, they are guaranteed not to be RAUW'ed. They effectively form a barrier between uniqued nodes, breaking some uniquing cycles, and shielding uniqued nodes from uniquing cycles. Unfortunately, with this barrier in place, the unresolved subgraph(s) can be disjoint from the top-level node. The mapping algorithm needs to find at least one representative from each disjoint subgraph. But which nodes are *stuck*, and which will get resolved automatically? And which nodes are in the unresolved subgraph? The old logic was conservative. This commit rotates the logic for distinct nodes, so that we have access to unresolved nodes at the top-level call to `llvm::MapMetadata()`. Each time we return to the top-level, we know that all temporaries have been RAUW'ed away. Here, it's safe (and necessary) to call `resolveCycles()` immediately on unresolved operands. This should also perform better than the old algorithm. The recursion stack is shorter, temporary nodes don't live as long, and there are fewer tracking references to unresolved nodes. As the debug info graph introduces more 'distinct' nodes, remapping should incrementally get cheaper and cheaper. Aside from possible performance improvements (and reduced cruft in the `LLVMContext`), there should be no functionality change here. llvm-svn: 244181
2015-08-06 01:52:42 +02:00
SmallVectorImpl<MDNode *> &DistinctWorklist,
ValueToValueMapTy &VM, RemapFlags Flags,
ValueMapTypeRemapper *TypeMapper,
ValueMaterializer *Materializer) {
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 the value already exists in the map, use it.
if (Metadata *NewMD = VM.MD().lookup(MD).get())
return NewMD;
if (isa<MDString>(MD))
return mapToSelf(VM, MD, Materializer, Flags);
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 (isa<ConstantAsMetadata>(MD))
if ((Flags & RF_NoModuleLevelChanges))
return mapToSelf(VM, MD, Materializer, Flags);
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 (const auto *VMD = dyn_cast<ValueAsMetadata>(MD)) {
Value *MappedV =
MapValue(VMD->getValue(), VM, Flags, TypeMapper, Materializer);
if (VMD->getValue() == MappedV ||
(!MappedV && (Flags & RF_IgnoreMissingEntries)))
return mapToSelf(VM, MD, Materializer, Flags);
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
// FIXME: This assert crashes during bootstrap, but I think it should be
// correct. For now, just match behaviour from before the metadata/value
// split.
//
Fix mapping of unmaterialized global values during metadata linking Summary: The patch to move metadata linking after global value linking didn't correctly map unmaterialized global values to null as desired. They were in fact mapped to the source copy. It largely worked by accident since most module linker clients destroyed the source module which caused the source GVs to be replaced by null, but caused a failure with LTO linking on Windows: http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20151109/312869.html The problem is that a null return value from materializeValueFor is handled by mapping the value to self. This is the desired behavior when materializeValueFor is passed a non-GlobalValue. The problem is how to distinguish that case from the case where we really do want to map to null. This patch addresses this by passing in a new flag to the value mapper indicating that unmapped global values should be mapped to null. Other Value types are handled as before. Note that the documented behavior of asserting on unmapped values when the flag RF_IgnoreMissingValues isn't set is currently disabled with FIXME notes due to bootstrap failures. I modified these disabled asserts so when they are eventually enabled again it won't assert for the unmapped values when the new RF_NullMapMissingGlobalValues flag is set. I also considered using a callback into the value materializer, but a flag seemed cleaner given that there are already existing flags. I also considered modifying materializeValueFor to return the input value when we want to map to source and then treat a null return to mean map to null. However, there are other value materializer subclasses that implement materializeValueFor, and they would all need to be audited and the return values possibly changed, which seemed error-prone. Reviewers: dexonsmith, joker.eph Subscribers: pcc, llvm-commits Differential Revision: http://reviews.llvm.org/D14682 llvm-svn: 253170
2015-11-15 15:50:14 +01:00
// assert((MappedV || (Flags & RF_NullMapMissingGlobalValues)) &&
// "Referenced metadata not in value map!");
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 (MappedV)
return mapToMetadata(VM, MD, ValueAsMetadata::get(MappedV), Materializer,
Flags);
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
return nullptr;
}
// Note: this cast precedes the Flags check so we always get its associated
// assertion.
const MDNode *Node = cast<MDNode>(MD);
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 is a module-level metadata and we know that nothing at the
// module level is changing, then use an identity mapping.
if (Flags & RF_NoModuleLevelChanges)
return mapToSelf(VM, MD, Materializer, Flags);
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
// Require resolved nodes whenever metadata might be remapped.
assert(((Flags & RF_HaveUnmaterializedMetadata) || Node->isResolved()) &&
"Unexpected unresolved node");
if (Materializer && Node->isTemporary()) {
assert(Flags & RF_HaveUnmaterializedMetadata);
Metadata *TempMD =
Materializer->mapTemporaryMetadata(const_cast<Metadata *>(MD));
// If the above callback returned an existing temporary node, use it
// instead of the current temporary node. This happens when earlier
// function importing passes already created and saved a temporary
// metadata node for the same value id.
if (TempMD) {
mapToMetadata(VM, MD, TempMD, Materializer, Flags);
return TempMD;
}
}
if (Node->isDistinct())
ValueMapper: Rotate distinct node remapping algorithm Rotate the algorithm for remapping distinct nodes in order to simplify how uniquing cycles get resolved. This removes some of the recursion, and, most importantly, exposes all uniquing cycles at the top-level. Besides being a little more efficient -- temporary MDNodes won't live as long -- the clearer logic should help protect against bugs like those fixed in r243961 and r243976. What are uniquing cycles? Why do they present challenges when remapping metadata? !0 = !{!1} !1 = !{!0} !0 and !1 form a simple uniquing cycle. When remapping from one metadata graph to another, every uniquing cycle gets "duplicated" through a dance: !0-temp = !{!1?} ; map(!0): clone !0, VM[!0] = !0-temp !1-temp = !{!0?} ; ..map(!1): clone !1, VM[!1] = !1-temp !1-temp = !{!0-temp} ; ..map(!1): remap !1's operands !2 = !{!0-temp} ; ..map(!1): uniquify: !1-temp => !2 !0-temp = !{!2} ; map(!0): remap !0's operands !3 = !{!2} ; map(!0): uniquify: !0-temp => !3 ; Result !2 = !{!3} !3 = !{!2} (In the two "uniquify" steps above, the operands of !X-temp are compared to the operands of !X. If they're the same, then !X-temp gets RAUW'ed to !X; if they're different, then !X-temp is promoted to a new unique node. The latter case always hits in for uniquing cycles, so we duplicate all the nodes involved.) Why is this a problem? Uniquable Metadata nodes that have temporary node as transitive operands keep RAUW support until the temporary nodes get finalized. With non-cycles, this happens automatically: when a uniquable node's count of unresolved operands drops to zero, it immediately sheds its own RAUW support (possibly triggering the same in any node that references it). However, uniquing cycles create a reference cycle, and uniqued nodes that transitively reference a uniquing cycle are "stuck" in an unresolved state until someone calls `MDNode::resolveCycles()` on a node in the unresolved subgraph. Distinct nodes should help here (and mostly do): since they aren't uniqued anywhere, they are guaranteed not to be RAUW'ed. They effectively form a barrier between uniqued nodes, breaking some uniquing cycles, and shielding uniqued nodes from uniquing cycles. Unfortunately, with this barrier in place, the unresolved subgraph(s) can be disjoint from the top-level node. The mapping algorithm needs to find at least one representative from each disjoint subgraph. But which nodes are *stuck*, and which will get resolved automatically? And which nodes are in the unresolved subgraph? The old logic was conservative. This commit rotates the logic for distinct nodes, so that we have access to unresolved nodes at the top-level call to `llvm::MapMetadata()`. Each time we return to the top-level, we know that all temporaries have been RAUW'ed away. Here, it's safe (and necessary) to call `resolveCycles()` immediately on unresolved operands. This should also perform better than the old algorithm. The recursion stack is shorter, temporary nodes don't live as long, and there are fewer tracking references to unresolved nodes. As the debug info graph introduces more 'distinct' nodes, remapping should incrementally get cheaper and cheaper. Aside from possible performance improvements (and reduced cruft in the `LLVMContext`), there should be no functionality change here. llvm-svn: 244181
2015-08-06 01:52:42 +02:00
return mapDistinctNode(Node, DistinctWorklist, VM, Flags, TypeMapper,
Materializer);
ValueMapper: Rotate distinct node remapping algorithm Rotate the algorithm for remapping distinct nodes in order to simplify how uniquing cycles get resolved. This removes some of the recursion, and, most importantly, exposes all uniquing cycles at the top-level. Besides being a little more efficient -- temporary MDNodes won't live as long -- the clearer logic should help protect against bugs like those fixed in r243961 and r243976. What are uniquing cycles? Why do they present challenges when remapping metadata? !0 = !{!1} !1 = !{!0} !0 and !1 form a simple uniquing cycle. When remapping from one metadata graph to another, every uniquing cycle gets "duplicated" through a dance: !0-temp = !{!1?} ; map(!0): clone !0, VM[!0] = !0-temp !1-temp = !{!0?} ; ..map(!1): clone !1, VM[!1] = !1-temp !1-temp = !{!0-temp} ; ..map(!1): remap !1's operands !2 = !{!0-temp} ; ..map(!1): uniquify: !1-temp => !2 !0-temp = !{!2} ; map(!0): remap !0's operands !3 = !{!2} ; map(!0): uniquify: !0-temp => !3 ; Result !2 = !{!3} !3 = !{!2} (In the two "uniquify" steps above, the operands of !X-temp are compared to the operands of !X. If they're the same, then !X-temp gets RAUW'ed to !X; if they're different, then !X-temp is promoted to a new unique node. The latter case always hits in for uniquing cycles, so we duplicate all the nodes involved.) Why is this a problem? Uniquable Metadata nodes that have temporary node as transitive operands keep RAUW support until the temporary nodes get finalized. With non-cycles, this happens automatically: when a uniquable node's count of unresolved operands drops to zero, it immediately sheds its own RAUW support (possibly triggering the same in any node that references it). However, uniquing cycles create a reference cycle, and uniqued nodes that transitively reference a uniquing cycle are "stuck" in an unresolved state until someone calls `MDNode::resolveCycles()` on a node in the unresolved subgraph. Distinct nodes should help here (and mostly do): since they aren't uniqued anywhere, they are guaranteed not to be RAUW'ed. They effectively form a barrier between uniqued nodes, breaking some uniquing cycles, and shielding uniqued nodes from uniquing cycles. Unfortunately, with this barrier in place, the unresolved subgraph(s) can be disjoint from the top-level node. The mapping algorithm needs to find at least one representative from each disjoint subgraph. But which nodes are *stuck*, and which will get resolved automatically? And which nodes are in the unresolved subgraph? The old logic was conservative. This commit rotates the logic for distinct nodes, so that we have access to unresolved nodes at the top-level call to `llvm::MapMetadata()`. Each time we return to the top-level, we know that all temporaries have been RAUW'ed away. Here, it's safe (and necessary) to call `resolveCycles()` immediately on unresolved operands. This should also perform better than the old algorithm. The recursion stack is shorter, temporary nodes don't live as long, and there are fewer tracking references to unresolved nodes. As the debug info graph introduces more 'distinct' nodes, remapping should incrementally get cheaper and cheaper. Aside from possible performance improvements (and reduced cruft in the `LLVMContext`), there should be no functionality change here. llvm-svn: 244181
2015-08-06 01:52:42 +02:00
return mapUniquedNode(Node, DistinctWorklist, VM, Flags, TypeMapper,
Materializer);
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
}
Metadata *llvm::MapMetadata(const Metadata *MD, ValueToValueMapTy &VM,
RemapFlags Flags, ValueMapTypeRemapper *TypeMapper,
ValueMaterializer *Materializer) {
ValueMapper: Rotate distinct node remapping algorithm Rotate the algorithm for remapping distinct nodes in order to simplify how uniquing cycles get resolved. This removes some of the recursion, and, most importantly, exposes all uniquing cycles at the top-level. Besides being a little more efficient -- temporary MDNodes won't live as long -- the clearer logic should help protect against bugs like those fixed in r243961 and r243976. What are uniquing cycles? Why do they present challenges when remapping metadata? !0 = !{!1} !1 = !{!0} !0 and !1 form a simple uniquing cycle. When remapping from one metadata graph to another, every uniquing cycle gets "duplicated" through a dance: !0-temp = !{!1?} ; map(!0): clone !0, VM[!0] = !0-temp !1-temp = !{!0?} ; ..map(!1): clone !1, VM[!1] = !1-temp !1-temp = !{!0-temp} ; ..map(!1): remap !1's operands !2 = !{!0-temp} ; ..map(!1): uniquify: !1-temp => !2 !0-temp = !{!2} ; map(!0): remap !0's operands !3 = !{!2} ; map(!0): uniquify: !0-temp => !3 ; Result !2 = !{!3} !3 = !{!2} (In the two "uniquify" steps above, the operands of !X-temp are compared to the operands of !X. If they're the same, then !X-temp gets RAUW'ed to !X; if they're different, then !X-temp is promoted to a new unique node. The latter case always hits in for uniquing cycles, so we duplicate all the nodes involved.) Why is this a problem? Uniquable Metadata nodes that have temporary node as transitive operands keep RAUW support until the temporary nodes get finalized. With non-cycles, this happens automatically: when a uniquable node's count of unresolved operands drops to zero, it immediately sheds its own RAUW support (possibly triggering the same in any node that references it). However, uniquing cycles create a reference cycle, and uniqued nodes that transitively reference a uniquing cycle are "stuck" in an unresolved state until someone calls `MDNode::resolveCycles()` on a node in the unresolved subgraph. Distinct nodes should help here (and mostly do): since they aren't uniqued anywhere, they are guaranteed not to be RAUW'ed. They effectively form a barrier between uniqued nodes, breaking some uniquing cycles, and shielding uniqued nodes from uniquing cycles. Unfortunately, with this barrier in place, the unresolved subgraph(s) can be disjoint from the top-level node. The mapping algorithm needs to find at least one representative from each disjoint subgraph. But which nodes are *stuck*, and which will get resolved automatically? And which nodes are in the unresolved subgraph? The old logic was conservative. This commit rotates the logic for distinct nodes, so that we have access to unresolved nodes at the top-level call to `llvm::MapMetadata()`. Each time we return to the top-level, we know that all temporaries have been RAUW'ed away. Here, it's safe (and necessary) to call `resolveCycles()` immediately on unresolved operands. This should also perform better than the old algorithm. The recursion stack is shorter, temporary nodes don't live as long, and there are fewer tracking references to unresolved nodes. As the debug info graph introduces more 'distinct' nodes, remapping should incrementally get cheaper and cheaper. Aside from possible performance improvements (and reduced cruft in the `LLVMContext`), there should be no functionality change here. llvm-svn: 244181
2015-08-06 01:52:42 +02:00
SmallVector<MDNode *, 8> DistinctWorklist;
Metadata *NewMD = MapMetadataImpl(MD, DistinctWorklist, VM, Flags, TypeMapper,
Materializer);
ValueMapper: Rotate distinct node remapping algorithm Rotate the algorithm for remapping distinct nodes in order to simplify how uniquing cycles get resolved. This removes some of the recursion, and, most importantly, exposes all uniquing cycles at the top-level. Besides being a little more efficient -- temporary MDNodes won't live as long -- the clearer logic should help protect against bugs like those fixed in r243961 and r243976. What are uniquing cycles? Why do they present challenges when remapping metadata? !0 = !{!1} !1 = !{!0} !0 and !1 form a simple uniquing cycle. When remapping from one metadata graph to another, every uniquing cycle gets "duplicated" through a dance: !0-temp = !{!1?} ; map(!0): clone !0, VM[!0] = !0-temp !1-temp = !{!0?} ; ..map(!1): clone !1, VM[!1] = !1-temp !1-temp = !{!0-temp} ; ..map(!1): remap !1's operands !2 = !{!0-temp} ; ..map(!1): uniquify: !1-temp => !2 !0-temp = !{!2} ; map(!0): remap !0's operands !3 = !{!2} ; map(!0): uniquify: !0-temp => !3 ; Result !2 = !{!3} !3 = !{!2} (In the two "uniquify" steps above, the operands of !X-temp are compared to the operands of !X. If they're the same, then !X-temp gets RAUW'ed to !X; if they're different, then !X-temp is promoted to a new unique node. The latter case always hits in for uniquing cycles, so we duplicate all the nodes involved.) Why is this a problem? Uniquable Metadata nodes that have temporary node as transitive operands keep RAUW support until the temporary nodes get finalized. With non-cycles, this happens automatically: when a uniquable node's count of unresolved operands drops to zero, it immediately sheds its own RAUW support (possibly triggering the same in any node that references it). However, uniquing cycles create a reference cycle, and uniqued nodes that transitively reference a uniquing cycle are "stuck" in an unresolved state until someone calls `MDNode::resolveCycles()` on a node in the unresolved subgraph. Distinct nodes should help here (and mostly do): since they aren't uniqued anywhere, they are guaranteed not to be RAUW'ed. They effectively form a barrier between uniqued nodes, breaking some uniquing cycles, and shielding uniqued nodes from uniquing cycles. Unfortunately, with this barrier in place, the unresolved subgraph(s) can be disjoint from the top-level node. The mapping algorithm needs to find at least one representative from each disjoint subgraph. But which nodes are *stuck*, and which will get resolved automatically? And which nodes are in the unresolved subgraph? The old logic was conservative. This commit rotates the logic for distinct nodes, so that we have access to unresolved nodes at the top-level call to `llvm::MapMetadata()`. Each time we return to the top-level, we know that all temporaries have been RAUW'ed away. Here, it's safe (and necessary) to call `resolveCycles()` immediately on unresolved operands. This should also perform better than the old algorithm. The recursion stack is shorter, temporary nodes don't live as long, and there are fewer tracking references to unresolved nodes. As the debug info graph introduces more 'distinct' nodes, remapping should incrementally get cheaper and cheaper. Aside from possible performance improvements (and reduced cruft in the `LLVMContext`), there should be no functionality change here. llvm-svn: 244181
2015-08-06 01:52:42 +02:00
// When there are no module-level changes, it's possible that the metadata
// graph has temporaries. Skip the logic to resolve cycles, since it's
// unnecessary (and invalid) in that case.
if (Flags & RF_NoModuleLevelChanges)
return NewMD;
// Resolve cycles involving the entry metadata.
resolveCycles(NewMD, Flags & RF_HaveUnmaterializedMetadata);
ValueMapper: Rotate distinct node remapping algorithm Rotate the algorithm for remapping distinct nodes in order to simplify how uniquing cycles get resolved. This removes some of the recursion, and, most importantly, exposes all uniquing cycles at the top-level. Besides being a little more efficient -- temporary MDNodes won't live as long -- the clearer logic should help protect against bugs like those fixed in r243961 and r243976. What are uniquing cycles? Why do they present challenges when remapping metadata? !0 = !{!1} !1 = !{!0} !0 and !1 form a simple uniquing cycle. When remapping from one metadata graph to another, every uniquing cycle gets "duplicated" through a dance: !0-temp = !{!1?} ; map(!0): clone !0, VM[!0] = !0-temp !1-temp = !{!0?} ; ..map(!1): clone !1, VM[!1] = !1-temp !1-temp = !{!0-temp} ; ..map(!1): remap !1's operands !2 = !{!0-temp} ; ..map(!1): uniquify: !1-temp => !2 !0-temp = !{!2} ; map(!0): remap !0's operands !3 = !{!2} ; map(!0): uniquify: !0-temp => !3 ; Result !2 = !{!3} !3 = !{!2} (In the two "uniquify" steps above, the operands of !X-temp are compared to the operands of !X. If they're the same, then !X-temp gets RAUW'ed to !X; if they're different, then !X-temp is promoted to a new unique node. The latter case always hits in for uniquing cycles, so we duplicate all the nodes involved.) Why is this a problem? Uniquable Metadata nodes that have temporary node as transitive operands keep RAUW support until the temporary nodes get finalized. With non-cycles, this happens automatically: when a uniquable node's count of unresolved operands drops to zero, it immediately sheds its own RAUW support (possibly triggering the same in any node that references it). However, uniquing cycles create a reference cycle, and uniqued nodes that transitively reference a uniquing cycle are "stuck" in an unresolved state until someone calls `MDNode::resolveCycles()` on a node in the unresolved subgraph. Distinct nodes should help here (and mostly do): since they aren't uniqued anywhere, they are guaranteed not to be RAUW'ed. They effectively form a barrier between uniqued nodes, breaking some uniquing cycles, and shielding uniqued nodes from uniquing cycles. Unfortunately, with this barrier in place, the unresolved subgraph(s) can be disjoint from the top-level node. The mapping algorithm needs to find at least one representative from each disjoint subgraph. But which nodes are *stuck*, and which will get resolved automatically? And which nodes are in the unresolved subgraph? The old logic was conservative. This commit rotates the logic for distinct nodes, so that we have access to unresolved nodes at the top-level call to `llvm::MapMetadata()`. Each time we return to the top-level, we know that all temporaries have been RAUW'ed away. Here, it's safe (and necessary) to call `resolveCycles()` immediately on unresolved operands. This should also perform better than the old algorithm. The recursion stack is shorter, temporary nodes don't live as long, and there are fewer tracking references to unresolved nodes. As the debug info graph introduces more 'distinct' nodes, remapping should incrementally get cheaper and cheaper. Aside from possible performance improvements (and reduced cruft in the `LLVMContext`), there should be no functionality change here. llvm-svn: 244181
2015-08-06 01:52:42 +02:00
// Remap the operands of distinct MDNodes.
while (!DistinctWorklist.empty())
remapOperands(*DistinctWorklist.pop_back_val(), DistinctWorklist, VM, Flags,
TypeMapper, Materializer);
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
return NewMD;
}
MDNode *llvm::MapMetadata(const MDNode *MD, ValueToValueMapTy &VM,
RemapFlags Flags, ValueMapTypeRemapper *TypeMapper,
ValueMaterializer *Materializer) {
return cast<MDNode>(MapMetadata(static_cast<const Metadata *>(MD), VM, Flags,
TypeMapper, Materializer));
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
}
/// RemapInstruction - Convert the instruction operands from referencing the
/// current values into those specified by VMap.
///
void llvm::RemapInstruction(Instruction *I, ValueToValueMapTy &VMap,
RemapFlags Flags, ValueMapTypeRemapper *TypeMapper,
ValueMaterializer *Materializer){
// Remap operands.
for (User::op_iterator op = I->op_begin(), E = I->op_end(); op != E; ++op) {
Value *V = MapValue(*op, VMap, Flags, TypeMapper, Materializer);
// If we aren't ignoring missing entries, assert that something happened.
if (V)
*op = V;
else
assert((Flags & RF_IgnoreMissingEntries) &&
"Referenced value not in value map!");
}
// Remap phi nodes' incoming blocks.
if (PHINode *PN = dyn_cast<PHINode>(I)) {
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
Value *V = MapValue(PN->getIncomingBlock(i), VMap, Flags);
// If we aren't ignoring missing entries, assert that something happened.
if (V)
PN->setIncomingBlock(i, cast<BasicBlock>(V));
else
assert((Flags & RF_IgnoreMissingEntries) &&
"Referenced block not in value map!");
}
}
// Remap attached metadata.
SmallVector<std::pair<unsigned, MDNode *>, 4> MDs;
I->getAllMetadata(MDs);
for (const auto &MI : MDs) {
MDNode *Old = MI.second;
MDNode *New = MapMetadata(Old, VMap, Flags, TypeMapper, Materializer);
if (New != Old)
I->setMetadata(MI.first, New);
}
if (!TypeMapper)
return;
// If the instruction's type is being remapped, do so now.
if (auto CS = CallSite(I)) {
SmallVector<Type *, 3> Tys;
FunctionType *FTy = CS.getFunctionType();
Tys.reserve(FTy->getNumParams());
for (Type *Ty : FTy->params())
Tys.push_back(TypeMapper->remapType(Ty));
CS.mutateFunctionType(FunctionType::get(
TypeMapper->remapType(I->getType()), Tys, FTy->isVarArg()));
return;
}
if (auto *AI = dyn_cast<AllocaInst>(I))
AI->setAllocatedType(TypeMapper->remapType(AI->getAllocatedType()));
if (auto *GEP = dyn_cast<GetElementPtrInst>(I)) {
GEP->setSourceElementType(
TypeMapper->remapType(GEP->getSourceElementType()));
GEP->setResultElementType(
TypeMapper->remapType(GEP->getResultElementType()));
}
I->mutateType(TypeMapper->remapType(I->getType()));
}