llvm-6502/include/llvm/IR/Metadata.h

1144 lines
37 KiB
C
Raw Normal View History

//===-- llvm/Metadata.h - Metadata definitions ------------------*- C++ -*-===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
/// @file
/// This file contains the declarations for metadata subclasses.
/// They represent the different flavors of metadata that live in LLVM.
//
//===----------------------------------------------------------------------===//
#ifndef LLVM_IR_METADATA_H
#define LLVM_IR_METADATA_H
#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/ilist_node.h"
#include "llvm/ADT/iterator_range.h"
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
#include "llvm/IR/Constant.h"
#include "llvm/IR/MetadataTracking.h"
#include "llvm/IR/Value.h"
#include "llvm/Support/ErrorHandling.h"
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
#include <type_traits>
namespace llvm {
class LLVMContext;
class Module;
template<typename ValueSubClass, typename ItemParentClass>
class SymbolTableListTraits;
enum LLVMConstants : uint32_t {
DEBUG_METADATA_VERSION = 2 // Current debug info version number.
};
/// \brief Root of the metadata hierarchy.
///
/// This is a root class for typeless data in the IR.
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
class Metadata {
friend class ReplaceableMetadataImpl;
/// \brief RTTI.
const unsigned char SubclassID;
protected:
/// \brief Active type of storage.
enum StorageType { Uniqued, Distinct, Temporary };
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
/// \brief Storage flag for non-uniqued, otherwise unowned, metadata.
StorageType Storage : 2;
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
// TODO: expose remaining bits to subclasses.
unsigned short SubclassData16;
unsigned SubclassData32;
public:
enum MetadataKind {
MDTupleKind,
MDLocationKind,
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
ConstantAsMetadataKind,
LocalAsMetadataKind,
MDStringKind
};
protected:
Metadata(unsigned ID, StorageType Storage)
: SubclassID(ID), Storage(Storage), SubclassData16(0), SubclassData32(0) {
}
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
~Metadata() {}
/// \brief Default handling of a changed operand, which asserts.
///
/// If subclasses pass themselves in as owners to a tracking node reference,
/// they must provide an implementation of this method.
void handleChangedOperand(void *, Metadata *) {
llvm_unreachable("Unimplemented in Metadata subclass");
}
public:
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
unsigned getMetadataID() const { return SubclassID; }
/// \brief User-friendly dump.
void dump() const;
void print(raw_ostream &OS) const;
void printAsOperand(raw_ostream &OS, bool PrintType = true,
const Module *M = nullptr) const;
};
#define HANDLE_METADATA(CLASS) class CLASS;
#include "llvm/IR/Metadata.def"
inline raw_ostream &operator<<(raw_ostream &OS, const Metadata &MD) {
MD.print(OS);
return OS;
}
/// \brief Metadata wrapper in the Value hierarchy.
///
/// A member of the \a Value hierarchy to represent a reference to metadata.
/// This allows, e.g., instrinsics to have metadata as operands.
///
/// Notably, this is the only thing in either hierarchy that is allowed to
/// reference \a LocalAsMetadata.
class MetadataAsValue : public Value {
friend class ReplaceableMetadataImpl;
friend class LLVMContextImpl;
Metadata *MD;
MetadataAsValue(Type *Ty, Metadata *MD);
~MetadataAsValue();
/// \brief Drop use of metadata (during teardown).
void dropUse() { MD = nullptr; }
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
public:
static MetadataAsValue *get(LLVMContext &Context, Metadata *MD);
static MetadataAsValue *getIfExists(LLVMContext &Context, Metadata *MD);
Metadata *getMetadata() const { return MD; }
static bool classof(const 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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
return V->getValueID() == MetadataAsValueVal;
}
private:
void handleChangedMetadata(Metadata *MD);
void track();
void untrack();
};
/// \brief Shared implementation of use-lists for replaceable metadata.
///
/// Most metadata cannot be RAUW'ed. This is a shared implementation of
/// use-lists and associated API for the two that support it (\a ValueAsMetadata
/// and \a TempMDNode).
class ReplaceableMetadataImpl {
friend class MetadataTracking;
public:
typedef MetadataTracking::OwnerTy OwnerTy;
private:
LLVMContext &Context;
uint64_t NextIndex;
SmallDenseMap<void *, std::pair<OwnerTy, uint64_t>, 4> UseMap;
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
public:
ReplaceableMetadataImpl(LLVMContext &Context)
: Context(Context), NextIndex(0) {}
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
~ReplaceableMetadataImpl() {
assert(UseMap.empty() && "Cannot destroy in-use replaceable metadata");
}
LLVMContext &getContext() const { return Context; }
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
/// \brief Replace all uses of this with MD.
///
/// Replace all uses of this with \c MD, which is allowed to be null.
void replaceAllUsesWith(Metadata *MD);
/// \brief Resolve all uses of this.
///
/// Resolve all uses of this, turning off RAUW permanently. If \c
/// ResolveUsers, call \a UniquableMDNode::resolve() on any users whose last
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
/// operand is resolved.
void resolveAllUses(bool ResolveUsers = true);
private:
void addRef(void *Ref, OwnerTy Owner);
void dropRef(void *Ref);
void moveRef(void *Ref, void *New, const Metadata &MD);
static ReplaceableMetadataImpl *get(Metadata &MD);
};
/// \brief Value wrapper in the Metadata hierarchy.
///
/// This is a custom value handle that allows other metadata to refer to
/// classes in the Value hierarchy.
///
/// Because of full uniquing support, each value is only wrapped by a single \a
/// ValueAsMetadata object, so the lookup maps are far more efficient than
/// those using ValueHandleBase.
class ValueAsMetadata : public Metadata, ReplaceableMetadataImpl {
friend class ReplaceableMetadataImpl;
friend class LLVMContextImpl;
Value *V;
/// \brief Drop users without RAUW (during teardown).
void dropUsers() {
ReplaceableMetadataImpl::resolveAllUses(/* ResolveUsers */ false);
}
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
protected:
ValueAsMetadata(unsigned ID, Value *V)
: Metadata(ID, Uniqued), ReplaceableMetadataImpl(V->getContext()), V(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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
assert(V && "Expected valid value");
}
~ValueAsMetadata() {}
public:
static ValueAsMetadata *get(Value *V);
static ConstantAsMetadata *getConstant(Value *C) {
return cast<ConstantAsMetadata>(get(C));
}
static LocalAsMetadata *getLocal(Value *Local) {
return cast<LocalAsMetadata>(get(Local));
}
static ValueAsMetadata *getIfExists(Value *V);
static ConstantAsMetadata *getConstantIfExists(Value *C) {
return cast_or_null<ConstantAsMetadata>(getIfExists(C));
}
static LocalAsMetadata *getLocalIfExists(Value *Local) {
return cast_or_null<LocalAsMetadata>(getIfExists(Local));
}
Value *getValue() const { return V; }
Type *getType() const { return V->getType(); }
LLVMContext &getContext() const { return V->getContext(); }
static void handleDeletion(Value *V);
static void handleRAUW(Value *From, Value *To);
protected:
/// \brief Handle collisions after \a Value::replaceAllUsesWith().
///
/// RAUW isn't supported directly for \a ValueAsMetadata, but if the wrapped
/// \a Value gets RAUW'ed and the target already exists, this is used to
/// merge the two metadata nodes.
void replaceAllUsesWith(Metadata *MD) {
ReplaceableMetadataImpl::replaceAllUsesWith(MD);
}
public:
static bool classof(const Metadata *MD) {
return MD->getMetadataID() == LocalAsMetadataKind ||
MD->getMetadataID() == ConstantAsMetadataKind;
}
};
class ConstantAsMetadata : public ValueAsMetadata {
friend class ValueAsMetadata;
ConstantAsMetadata(Constant *C)
: ValueAsMetadata(ConstantAsMetadataKind, C) {}
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
public:
static ConstantAsMetadata *get(Constant *C) {
return ValueAsMetadata::getConstant(C);
}
static ConstantAsMetadata *getIfExists(Constant *C) {
return ValueAsMetadata::getConstantIfExists(C);
}
Constant *getValue() const {
return cast<Constant>(ValueAsMetadata::getValue());
}
static bool classof(const Metadata *MD) {
return MD->getMetadataID() == ConstantAsMetadataKind;
}
};
class LocalAsMetadata : public ValueAsMetadata {
friend class ValueAsMetadata;
LocalAsMetadata(Value *Local)
: ValueAsMetadata(LocalAsMetadataKind, Local) {
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
assert(!isa<Constant>(Local) && "Expected local value");
}
public:
static LocalAsMetadata *get(Value *Local) {
return ValueAsMetadata::getLocal(Local);
}
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
static LocalAsMetadata *getIfExists(Value *Local) {
return ValueAsMetadata::getLocalIfExists(Local);
}
static bool classof(const Metadata *MD) {
return MD->getMetadataID() == LocalAsMetadataKind;
}
};
/// \brief Transitional API for extracting constants from Metadata.
///
/// This namespace contains transitional functions for metadata that points to
/// \a Constants.
///
/// In prehistory -- when metadata was a subclass of \a Value -- \a MDNode
/// operands could refer to any \a Value. There's was a lot of code like this:
///
/// \code
/// MDNode *N = ...;
/// auto *CI = dyn_cast<ConstantInt>(N->getOperand(2));
/// \endcode
///
/// Now that \a Value and \a Metadata are in separate hierarchies, maintaining
/// the semantics for \a isa(), \a cast(), \a dyn_cast() (etc.) requires three
/// steps: cast in the \a Metadata hierarchy, extraction of the \a Value, and
/// cast in the \a Value hierarchy. Besides creating boiler-plate, this
/// requires subtle control flow changes.
///
/// The end-goal is to create a new type of metadata, called (e.g.) \a MDInt,
/// so that metadata can refer to numbers without traversing a bridge to the \a
/// Value hierarchy. In this final state, the code above would look like this:
///
/// \code
/// MDNode *N = ...;
/// auto *MI = dyn_cast<MDInt>(N->getOperand(2));
/// \endcode
///
/// The API in this namespace supports the transition. \a MDInt doesn't exist
/// yet, and even once it does, changing each metadata schema to use it is its
/// own mini-project. In the meantime this API prevents us from introducing
/// complex and bug-prone control flow that will disappear in the end. In
/// particular, the above code looks like this:
///
/// \code
/// MDNode *N = ...;
/// auto *CI = mdconst::dyn_extract<ConstantInt>(N->getOperand(2));
/// \endcode
///
/// The full set of provided functions includes:
///
/// mdconst::hasa <=> isa
/// mdconst::extract <=> cast
/// mdconst::extract_or_null <=> cast_or_null
/// mdconst::dyn_extract <=> dyn_cast
/// mdconst::dyn_extract_or_null <=> dyn_cast_or_null
///
/// The target of the cast must be a subclass of \a Constant.
namespace mdconst {
namespace detail {
template <class T> T &make();
template <class T, class Result> struct HasDereference {
typedef char Yes[1];
typedef char No[2];
template <size_t N> struct SFINAE {};
template <class U, class V>
static Yes &hasDereference(SFINAE<sizeof(static_cast<V>(*make<U>()))> * = 0);
template <class U, class V> static No &hasDereference(...);
static const bool value =
sizeof(hasDereference<T, Result>(nullptr)) == sizeof(Yes);
};
template <class V, class M> struct IsValidPointer {
static const bool value = std::is_base_of<Constant, V>::value &&
HasDereference<M, const Metadata &>::value;
};
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
template <class V, class M> struct IsValidReference {
static const bool value = std::is_base_of<Constant, V>::value &&
std::is_convertible<M, const Metadata &>::value;
};
} // end namespace detail
/// \brief Check whether Metadata has a Value.
///
/// As an analogue to \a isa(), check whether \c MD has an \a Value inside of
/// type \c X.
template <class X, class Y>
inline typename std::enable_if<detail::IsValidPointer<X, Y>::value, bool>::type
hasa(Y &&MD) {
assert(MD && "Null pointer sent into hasa");
if (auto *V = dyn_cast<ConstantAsMetadata>(MD))
return isa<X>(V->getValue());
return false;
}
template <class X, class Y>
inline
typename std::enable_if<detail::IsValidReference<X, Y &>::value, bool>::type
hasa(Y &MD) {
return hasa(&MD);
}
/// \brief Extract a Value from Metadata.
///
/// As an analogue to \a cast(), extract the \a Value subclass \c X from \c MD.
template <class X, class Y>
inline typename std::enable_if<detail::IsValidPointer<X, Y>::value, X *>::type
extract(Y &&MD) {
return cast<X>(cast<ConstantAsMetadata>(MD)->getValue());
}
template <class X, class Y>
inline
typename std::enable_if<detail::IsValidReference<X, Y &>::value, X *>::type
extract(Y &MD) {
return extract(&MD);
}
/// \brief Extract a Value from Metadata, allowing null.
///
/// As an analogue to \a cast_or_null(), extract the \a Value subclass \c X
/// from \c MD, allowing \c MD to be null.
template <class X, class Y>
inline typename std::enable_if<detail::IsValidPointer<X, Y>::value, X *>::type
extract_or_null(Y &&MD) {
if (auto *V = cast_or_null<ConstantAsMetadata>(MD))
return cast<X>(V->getValue());
return nullptr;
}
/// \brief Extract a Value from Metadata, if any.
///
/// As an analogue to \a dyn_cast_or_null(), extract the \a Value subclass \c X
/// from \c MD, return null if \c MD doesn't contain a \a Value or if the \a
/// Value it does contain is of the wrong subclass.
template <class X, class Y>
inline typename std::enable_if<detail::IsValidPointer<X, Y>::value, X *>::type
dyn_extract(Y &&MD) {
if (auto *V = dyn_cast<ConstantAsMetadata>(MD))
return dyn_cast<X>(V->getValue());
return nullptr;
}
/// \brief Extract a Value from Metadata, if any, allowing null.
///
/// As an analogue to \a dyn_cast_or_null(), extract the \a Value subclass \c X
/// from \c MD, return null if \c MD doesn't contain a \a Value or if the \a
/// Value it does contain is of the wrong subclass, allowing \c MD to be null.
template <class X, class Y>
inline typename std::enable_if<detail::IsValidPointer<X, Y>::value, X *>::type
dyn_extract_or_null(Y &&MD) {
if (auto *V = dyn_cast_or_null<ConstantAsMetadata>(MD))
return dyn_cast<X>(V->getValue());
return nullptr;
}
} // end namespace mdconst
//===----------------------------------------------------------------------===//
/// \brief A single uniqued string.
///
/// These are used to efficiently contain a byte sequence for metadata.
/// MDString is always unnamed.
class MDString : public Metadata {
friend class StringMapEntry<MDString>;
MDString(const MDString &) LLVM_DELETED_FUNCTION;
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
MDString &operator=(MDString &&) LLVM_DELETED_FUNCTION;
MDString &operator=(const MDString &) LLVM_DELETED_FUNCTION;
StringMapEntry<MDString> *Entry;
MDString() : Metadata(MDStringKind, Uniqued), Entry(nullptr) {}
MDString(MDString &&) : Metadata(MDStringKind, Uniqued) {}
public:
static MDString *get(LLVMContext &Context, StringRef Str);
static MDString *get(LLVMContext &Context, const char *Str) {
return get(Context, Str ? StringRef(Str) : StringRef());
}
StringRef getString() const;
unsigned getLength() const { return (unsigned)getString().size(); }
typedef StringRef::iterator iterator;
/// \brief Pointer to the first byte of the string.
iterator begin() const { return getString().begin(); }
/// \brief Pointer to one byte past the end of the string.
iterator end() const { return getString().end(); }
const unsigned char *bytes_begin() const { return getString().bytes_begin(); }
const unsigned char *bytes_end() const { return getString().bytes_end(); }
/// \brief Methods for support type inquiry through isa, cast, and dyn_cast.
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
static bool classof(const Metadata *MD) {
return MD->getMetadataID() == MDStringKind;
}
};
/// \brief A collection of metadata nodes that might be associated with a
/// memory access used by the alias-analysis infrastructure.
struct AAMDNodes {
explicit AAMDNodes(MDNode *T = nullptr, MDNode *S = nullptr,
MDNode *N = nullptr)
: TBAA(T), Scope(S), NoAlias(N) {}
bool operator==(const AAMDNodes &A) const {
return TBAA == A.TBAA && Scope == A.Scope && NoAlias == A.NoAlias;
}
bool operator!=(const AAMDNodes &A) const { return !(*this == A); }
LLVM_EXPLICIT operator bool() const { return TBAA || Scope || NoAlias; }
/// \brief The tag for type-based alias analysis.
MDNode *TBAA;
/// \brief The tag for alias scope specification (used with noalias).
Add scoped-noalias metadata This commit adds scoped noalias metadata. The primary motivations for this feature are: 1. To preserve noalias function attribute information when inlining 2. To provide the ability to model block-scope C99 restrict pointers Neither of these two abilities are added here, only the necessary infrastructure. In fact, there should be no change to existing functionality, only the addition of new features. The logic that converts noalias function parameters into this metadata during inlining will come in a follow-up commit. What is added here is the ability to generally specify noalias memory-access sets. Regarding the metadata, alias-analysis scopes are defined similar to TBAA nodes: !scope0 = metadata !{ metadata !"scope of foo()" } !scope1 = metadata !{ metadata !"scope 1", metadata !scope0 } !scope2 = metadata !{ metadata !"scope 2", metadata !scope0 } !scope3 = metadata !{ metadata !"scope 2.1", metadata !scope2 } !scope4 = metadata !{ metadata !"scope 2.2", metadata !scope2 } Loads and stores can be tagged with an alias-analysis scope, and also, with a noalias tag for a specific scope: ... = load %ptr1, !alias.scope !{ !scope1 } ... = load %ptr2, !alias.scope !{ !scope1, !scope2 }, !noalias !{ !scope1 } When evaluating an aliasing query, if one of the instructions is associated with an alias.scope id that is identical to the noalias scope associated with the other instruction, or is a descendant (in the scope hierarchy) of the noalias scope associated with the other instruction, then the two memory accesses are assumed not to alias. Note that is the first element of the scope metadata is a string, then it can be combined accross functions and translation units. The string can be replaced by a self-reference to create globally unqiue scope identifiers. [Note: This overview is slightly stylized, since the metadata nodes really need to just be numbers (!0 instead of !scope0), and the scope lists are also global unnamed metadata.] Existing noalias metadata in a callee is "cloned" for use by the inlined code. This is necessary because the aliasing scopes are unique to each call site (because of possible control dependencies on the aliasing properties). For example, consider a function: foo(noalias a, noalias b) { *a = *b; } that gets inlined into bar() { ... if (...) foo(a1, b1); ... if (...) foo(a2, b2); } -- now just because we know that a1 does not alias with b1 at the first call site, and a2 does not alias with b2 at the second call site, we cannot let inlining these functons have the metadata imply that a1 does not alias with b2. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@213864 91177308-0d34-0410-b5e6-96231b3b80d8
2014-07-24 14:25:39 +00:00
MDNode *Scope;
/// \brief The tag specifying the noalias scope.
Add scoped-noalias metadata This commit adds scoped noalias metadata. The primary motivations for this feature are: 1. To preserve noalias function attribute information when inlining 2. To provide the ability to model block-scope C99 restrict pointers Neither of these two abilities are added here, only the necessary infrastructure. In fact, there should be no change to existing functionality, only the addition of new features. The logic that converts noalias function parameters into this metadata during inlining will come in a follow-up commit. What is added here is the ability to generally specify noalias memory-access sets. Regarding the metadata, alias-analysis scopes are defined similar to TBAA nodes: !scope0 = metadata !{ metadata !"scope of foo()" } !scope1 = metadata !{ metadata !"scope 1", metadata !scope0 } !scope2 = metadata !{ metadata !"scope 2", metadata !scope0 } !scope3 = metadata !{ metadata !"scope 2.1", metadata !scope2 } !scope4 = metadata !{ metadata !"scope 2.2", metadata !scope2 } Loads and stores can be tagged with an alias-analysis scope, and also, with a noalias tag for a specific scope: ... = load %ptr1, !alias.scope !{ !scope1 } ... = load %ptr2, !alias.scope !{ !scope1, !scope2 }, !noalias !{ !scope1 } When evaluating an aliasing query, if one of the instructions is associated with an alias.scope id that is identical to the noalias scope associated with the other instruction, or is a descendant (in the scope hierarchy) of the noalias scope associated with the other instruction, then the two memory accesses are assumed not to alias. Note that is the first element of the scope metadata is a string, then it can be combined accross functions and translation units. The string can be replaced by a self-reference to create globally unqiue scope identifiers. [Note: This overview is slightly stylized, since the metadata nodes really need to just be numbers (!0 instead of !scope0), and the scope lists are also global unnamed metadata.] Existing noalias metadata in a callee is "cloned" for use by the inlined code. This is necessary because the aliasing scopes are unique to each call site (because of possible control dependencies on the aliasing properties). For example, consider a function: foo(noalias a, noalias b) { *a = *b; } that gets inlined into bar() { ... if (...) foo(a1, b1); ... if (...) foo(a2, b2); } -- now just because we know that a1 does not alias with b1 at the first call site, and a2 does not alias with b2 at the second call site, we cannot let inlining these functons have the metadata imply that a1 does not alias with b2. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@213864 91177308-0d34-0410-b5e6-96231b3b80d8
2014-07-24 14:25:39 +00:00
MDNode *NoAlias;
};
// Specialize DenseMapInfo for AAMDNodes.
template<>
struct DenseMapInfo<AAMDNodes> {
static inline AAMDNodes getEmptyKey() {
Add scoped-noalias metadata This commit adds scoped noalias metadata. The primary motivations for this feature are: 1. To preserve noalias function attribute information when inlining 2. To provide the ability to model block-scope C99 restrict pointers Neither of these two abilities are added here, only the necessary infrastructure. In fact, there should be no change to existing functionality, only the addition of new features. The logic that converts noalias function parameters into this metadata during inlining will come in a follow-up commit. What is added here is the ability to generally specify noalias memory-access sets. Regarding the metadata, alias-analysis scopes are defined similar to TBAA nodes: !scope0 = metadata !{ metadata !"scope of foo()" } !scope1 = metadata !{ metadata !"scope 1", metadata !scope0 } !scope2 = metadata !{ metadata !"scope 2", metadata !scope0 } !scope3 = metadata !{ metadata !"scope 2.1", metadata !scope2 } !scope4 = metadata !{ metadata !"scope 2.2", metadata !scope2 } Loads and stores can be tagged with an alias-analysis scope, and also, with a noalias tag for a specific scope: ... = load %ptr1, !alias.scope !{ !scope1 } ... = load %ptr2, !alias.scope !{ !scope1, !scope2 }, !noalias !{ !scope1 } When evaluating an aliasing query, if one of the instructions is associated with an alias.scope id that is identical to the noalias scope associated with the other instruction, or is a descendant (in the scope hierarchy) of the noalias scope associated with the other instruction, then the two memory accesses are assumed not to alias. Note that is the first element of the scope metadata is a string, then it can be combined accross functions and translation units. The string can be replaced by a self-reference to create globally unqiue scope identifiers. [Note: This overview is slightly stylized, since the metadata nodes really need to just be numbers (!0 instead of !scope0), and the scope lists are also global unnamed metadata.] Existing noalias metadata in a callee is "cloned" for use by the inlined code. This is necessary because the aliasing scopes are unique to each call site (because of possible control dependencies on the aliasing properties). For example, consider a function: foo(noalias a, noalias b) { *a = *b; } that gets inlined into bar() { ... if (...) foo(a1, b1); ... if (...) foo(a2, b2); } -- now just because we know that a1 does not alias with b1 at the first call site, and a2 does not alias with b2 at the second call site, we cannot let inlining these functons have the metadata imply that a1 does not alias with b2. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@213864 91177308-0d34-0410-b5e6-96231b3b80d8
2014-07-24 14:25:39 +00:00
return AAMDNodes(DenseMapInfo<MDNode *>::getEmptyKey(), 0, 0);
}
static inline AAMDNodes getTombstoneKey() {
Add scoped-noalias metadata This commit adds scoped noalias metadata. The primary motivations for this feature are: 1. To preserve noalias function attribute information when inlining 2. To provide the ability to model block-scope C99 restrict pointers Neither of these two abilities are added here, only the necessary infrastructure. In fact, there should be no change to existing functionality, only the addition of new features. The logic that converts noalias function parameters into this metadata during inlining will come in a follow-up commit. What is added here is the ability to generally specify noalias memory-access sets. Regarding the metadata, alias-analysis scopes are defined similar to TBAA nodes: !scope0 = metadata !{ metadata !"scope of foo()" } !scope1 = metadata !{ metadata !"scope 1", metadata !scope0 } !scope2 = metadata !{ metadata !"scope 2", metadata !scope0 } !scope3 = metadata !{ metadata !"scope 2.1", metadata !scope2 } !scope4 = metadata !{ metadata !"scope 2.2", metadata !scope2 } Loads and stores can be tagged with an alias-analysis scope, and also, with a noalias tag for a specific scope: ... = load %ptr1, !alias.scope !{ !scope1 } ... = load %ptr2, !alias.scope !{ !scope1, !scope2 }, !noalias !{ !scope1 } When evaluating an aliasing query, if one of the instructions is associated with an alias.scope id that is identical to the noalias scope associated with the other instruction, or is a descendant (in the scope hierarchy) of the noalias scope associated with the other instruction, then the two memory accesses are assumed not to alias. Note that is the first element of the scope metadata is a string, then it can be combined accross functions and translation units. The string can be replaced by a self-reference to create globally unqiue scope identifiers. [Note: This overview is slightly stylized, since the metadata nodes really need to just be numbers (!0 instead of !scope0), and the scope lists are also global unnamed metadata.] Existing noalias metadata in a callee is "cloned" for use by the inlined code. This is necessary because the aliasing scopes are unique to each call site (because of possible control dependencies on the aliasing properties). For example, consider a function: foo(noalias a, noalias b) { *a = *b; } that gets inlined into bar() { ... if (...) foo(a1, b1); ... if (...) foo(a2, b2); } -- now just because we know that a1 does not alias with b1 at the first call site, and a2 does not alias with b2 at the second call site, we cannot let inlining these functons have the metadata imply that a1 does not alias with b2. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@213864 91177308-0d34-0410-b5e6-96231b3b80d8
2014-07-24 14:25:39 +00:00
return AAMDNodes(DenseMapInfo<MDNode *>::getTombstoneKey(), 0, 0);
}
static unsigned getHashValue(const AAMDNodes &Val) {
Add scoped-noalias metadata This commit adds scoped noalias metadata. The primary motivations for this feature are: 1. To preserve noalias function attribute information when inlining 2. To provide the ability to model block-scope C99 restrict pointers Neither of these two abilities are added here, only the necessary infrastructure. In fact, there should be no change to existing functionality, only the addition of new features. The logic that converts noalias function parameters into this metadata during inlining will come in a follow-up commit. What is added here is the ability to generally specify noalias memory-access sets. Regarding the metadata, alias-analysis scopes are defined similar to TBAA nodes: !scope0 = metadata !{ metadata !"scope of foo()" } !scope1 = metadata !{ metadata !"scope 1", metadata !scope0 } !scope2 = metadata !{ metadata !"scope 2", metadata !scope0 } !scope3 = metadata !{ metadata !"scope 2.1", metadata !scope2 } !scope4 = metadata !{ metadata !"scope 2.2", metadata !scope2 } Loads and stores can be tagged with an alias-analysis scope, and also, with a noalias tag for a specific scope: ... = load %ptr1, !alias.scope !{ !scope1 } ... = load %ptr2, !alias.scope !{ !scope1, !scope2 }, !noalias !{ !scope1 } When evaluating an aliasing query, if one of the instructions is associated with an alias.scope id that is identical to the noalias scope associated with the other instruction, or is a descendant (in the scope hierarchy) of the noalias scope associated with the other instruction, then the two memory accesses are assumed not to alias. Note that is the first element of the scope metadata is a string, then it can be combined accross functions and translation units. The string can be replaced by a self-reference to create globally unqiue scope identifiers. [Note: This overview is slightly stylized, since the metadata nodes really need to just be numbers (!0 instead of !scope0), and the scope lists are also global unnamed metadata.] Existing noalias metadata in a callee is "cloned" for use by the inlined code. This is necessary because the aliasing scopes are unique to each call site (because of possible control dependencies on the aliasing properties). For example, consider a function: foo(noalias a, noalias b) { *a = *b; } that gets inlined into bar() { ... if (...) foo(a1, b1); ... if (...) foo(a2, b2); } -- now just because we know that a1 does not alias with b1 at the first call site, and a2 does not alias with b2 at the second call site, we cannot let inlining these functons have the metadata imply that a1 does not alias with b2. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@213864 91177308-0d34-0410-b5e6-96231b3b80d8
2014-07-24 14:25:39 +00:00
return DenseMapInfo<MDNode *>::getHashValue(Val.TBAA) ^
DenseMapInfo<MDNode *>::getHashValue(Val.Scope) ^
DenseMapInfo<MDNode *>::getHashValue(Val.NoAlias);
}
static bool isEqual(const AAMDNodes &LHS, const AAMDNodes &RHS) {
return LHS == RHS;
}
};
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
/// \brief Tracking metadata reference owned by Metadata.
///
/// Similar to \a TrackingMDRef, but it's expected to be owned by an instance
/// of \a Metadata, which has the option of registering itself for callbacks to
/// re-unique itself.
///
/// In particular, this is used by \a MDNode.
class MDOperand {
MDOperand(MDOperand &&) LLVM_DELETED_FUNCTION;
MDOperand(const MDOperand &) LLVM_DELETED_FUNCTION;
MDOperand &operator=(MDOperand &&) LLVM_DELETED_FUNCTION;
MDOperand &operator=(const MDOperand &) LLVM_DELETED_FUNCTION;
Metadata *MD;
public:
MDOperand() : MD(nullptr) {}
~MDOperand() { untrack(); }
Metadata *get() const { return MD; }
operator Metadata *() const { return get(); }
Metadata *operator->() const { return get(); }
Metadata &operator*() const { return *get(); }
void reset() {
untrack();
MD = nullptr;
}
void reset(Metadata *MD, Metadata *Owner) {
untrack();
this->MD = MD;
track(Owner);
}
private:
void track(Metadata *Owner) {
if (MD) {
if (Owner)
MetadataTracking::track(this, *MD, *Owner);
else
MetadataTracking::track(MD);
}
}
void untrack() {
assert(static_cast<void *>(this) == &MD && "Expected same address");
if (MD)
MetadataTracking::untrack(MD);
}
};
template <> struct simplify_type<MDOperand> {
typedef Metadata *SimpleType;
static SimpleType getSimplifiedValue(MDOperand &MD) { return MD.get(); }
};
template <> struct simplify_type<const MDOperand> {
typedef Metadata *SimpleType;
static SimpleType getSimplifiedValue(const MDOperand &MD) { return MD.get(); }
};
/// \brief Pointer to the context, with optional RAUW support.
///
/// Either a raw (non-null) pointer to the \a LLVMContext, or an owned pointer
/// to \a ReplaceableMetadataImpl (which has a reference to \a LLVMContext).
class ContextAndReplaceableUses {
PointerUnion<LLVMContext *, ReplaceableMetadataImpl *> Ptr;
ContextAndReplaceableUses() LLVM_DELETED_FUNCTION;
ContextAndReplaceableUses(ContextAndReplaceableUses &&)
LLVM_DELETED_FUNCTION;
ContextAndReplaceableUses(const ContextAndReplaceableUses &)
LLVM_DELETED_FUNCTION;
ContextAndReplaceableUses &
operator=(ContextAndReplaceableUses &&) LLVM_DELETED_FUNCTION;
ContextAndReplaceableUses &
operator=(const ContextAndReplaceableUses &) LLVM_DELETED_FUNCTION;
public:
ContextAndReplaceableUses(LLVMContext &Context) : Ptr(&Context) {}
ContextAndReplaceableUses(
std::unique_ptr<ReplaceableMetadataImpl> ReplaceableUses)
: Ptr(ReplaceableUses.release()) {
assert(getReplaceableUses() && "Expected non-null replaceable uses");
}
~ContextAndReplaceableUses() { delete getReplaceableUses(); }
operator LLVMContext &() { return getContext(); }
/// \brief Whether this contains RAUW support.
bool hasReplaceableUses() const {
return Ptr.is<ReplaceableMetadataImpl *>();
}
LLVMContext &getContext() const {
if (hasReplaceableUses())
return getReplaceableUses()->getContext();
return *Ptr.get<LLVMContext *>();
}
ReplaceableMetadataImpl *getReplaceableUses() const {
if (hasReplaceableUses())
return Ptr.get<ReplaceableMetadataImpl *>();
return nullptr;
}
/// \brief Assign RAUW support to this.
///
/// Make this replaceable, taking ownership of \c ReplaceableUses (which must
/// not be null).
void
makeReplaceable(std::unique_ptr<ReplaceableMetadataImpl> ReplaceableUses) {
assert(ReplaceableUses && "Expected non-null replaceable uses");
assert(&ReplaceableUses->getContext() == &getContext() &&
"Expected same context");
delete getReplaceableUses();
Ptr = ReplaceableUses.release();
}
/// \brief Drop RAUW support.
///
/// Cede ownership of RAUW support, returning it.
std::unique_ptr<ReplaceableMetadataImpl> takeReplaceableUses() {
assert(hasReplaceableUses() && "Expected to own replaceable uses");
std::unique_ptr<ReplaceableMetadataImpl> ReplaceableUses(
getReplaceableUses());
Ptr = &ReplaceableUses->getContext();
return ReplaceableUses;
}
};
struct TempMDNodeDeleter {
inline void operator()(MDNode *Node) const;
};
#define HANDLE_UNIQUABLE_LEAF(CLASS) \
typedef std::unique_ptr<CLASS, TempMDNodeDeleter> Temp##CLASS;
#define HANDLE_UNIQUABLE_BRANCH(CLASS) HANDLE_UNIQUABLE_LEAF(CLASS)
#include "llvm/IR/Metadata.def"
//===----------------------------------------------------------------------===//
/// \brief Tuple of metadata.
class MDNode : public Metadata {
friend class ReplaceableMetadataImpl;
MDNode(const MDNode &) LLVM_DELETED_FUNCTION;
void operator=(const MDNode &) LLVM_DELETED_FUNCTION;
void *operator new(size_t) LLVM_DELETED_FUNCTION;
protected:
ContextAndReplaceableUses Context;
private:
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
unsigned NumOperands;
protected:
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
unsigned MDNodeSubclassData;
void *operator new(size_t Size, unsigned NumOps);
void operator delete(void *Mem);
/// \brief Required by std, but never called.
void operator delete(void *, unsigned) {
llvm_unreachable("Constructor throws?");
}
/// \brief Required by std, but never called.
void operator delete(void *, unsigned, bool) {
llvm_unreachable("Constructor throws?");
}
MDNode(LLVMContext &Context, unsigned ID, StorageType Storage,
ArrayRef<Metadata *> MDs);
~MDNode() {}
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
void dropAllReferences();
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
MDOperand *mutable_begin() { return mutable_end() - NumOperands; }
MDOperand *mutable_end() { return reinterpret_cast<MDOperand *>(this); }
public:
static inline MDTuple *get(LLVMContext &Context, ArrayRef<Metadata *> MDs);
static inline MDTuple *getIfExists(LLVMContext &Context,
ArrayRef<Metadata *> MDs);
static inline MDTuple *getDistinct(LLVMContext &Context,
ArrayRef<Metadata *> MDs);
static inline TempMDTuple getTemporary(LLVMContext &Context,
ArrayRef<Metadata *> MDs);
/// \brief Deallocate a node created by getTemporary.
///
/// The node must not have any users.
static void deleteTemporary(MDNode *N);
LLVMContext &getContext() const { return Context.getContext(); }
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
/// \brief Replace a specific operand.
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
void replaceOperandWith(unsigned I, Metadata *New);
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
/// \brief Check if node is fully resolved.
///
/// If \a isTemporary(), this always returns \c false; if \a isDistinct(),
/// this always returns \c true.
///
/// If \a isUniqued(), returns \c true if this has already dropped RAUW
/// support (because all operands are resolved).
///
/// As forward declarations are resolved, their containers should get
/// resolved automatically. However, if this (or one of its operands) is
/// involved in a cycle, \a resolveCycles() needs to be called explicitly.
bool isResolved() const { return !Context.hasReplaceableUses(); }
bool isUniqued() const { return Storage == Uniqued; }
bool isDistinct() const { return Storage == Distinct; }
bool isTemporary() const { return Storage == Temporary; }
/// \brief RAUW a temporary.
///
/// \pre \a isTemporary() must be \c true.
void replaceAllUsesWith(Metadata *MD) {
assert(isTemporary() && "Expected temporary node");
assert(!isResolved() && "Expected RAUW support");
Context.getReplaceableUses()->replaceAllUsesWith(MD);
}
/// \brief Replace a temporary node with a uniqued one.
///
/// Create a uniqued version of \c N -- in place, if possible -- and return
/// it. Takes ownership of the temporary node.
template <class T>
static typename std::enable_if<std::is_base_of<UniquableMDNode, T>::value,
T *>::type
replaceWithUniqued(std::unique_ptr<T, TempMDNodeDeleter> N);
/// \brief Replace a temporary node with a distinct one.
///
/// Create a distinct version of \c N -- in place, if possible -- and return
/// it. Takes ownership of the temporary node.
template <class T>
static typename std::enable_if<std::is_base_of<UniquableMDNode, T>::value,
T *>::type
replaceWithDistinct(std::unique_ptr<T, TempMDNodeDeleter> N);
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
protected:
/// \brief Set an operand.
///
/// Sets the operand directly, without worrying about uniquing.
void setOperand(unsigned I, Metadata *New);
public:
typedef const MDOperand *op_iterator;
typedef iterator_range<op_iterator> op_range;
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
op_iterator op_begin() const {
return const_cast<MDNode *>(this)->mutable_begin();
}
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
op_iterator op_end() const {
return const_cast<MDNode *>(this)->mutable_end();
}
op_range operands() const { return op_range(op_begin(), op_end()); }
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
const MDOperand &getOperand(unsigned I) const {
assert(I < NumOperands && "Out of range");
return op_begin()[I];
}
/// \brief Return number of MDNode operands.
unsigned getNumOperands() const { return NumOperands; }
/// \brief Methods for support type inquiry through isa, cast, and dyn_cast:
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
static bool classof(const Metadata *MD) {
return MD->getMetadataID() == MDTupleKind ||
MD->getMetadataID() == MDLocationKind;
}
/// \brief Check whether MDNode is a vtable access.
bool isTBAAVtableAccess() const;
/// \brief Methods for metadata merging.
Add scoped-noalias metadata This commit adds scoped noalias metadata. The primary motivations for this feature are: 1. To preserve noalias function attribute information when inlining 2. To provide the ability to model block-scope C99 restrict pointers Neither of these two abilities are added here, only the necessary infrastructure. In fact, there should be no change to existing functionality, only the addition of new features. The logic that converts noalias function parameters into this metadata during inlining will come in a follow-up commit. What is added here is the ability to generally specify noalias memory-access sets. Regarding the metadata, alias-analysis scopes are defined similar to TBAA nodes: !scope0 = metadata !{ metadata !"scope of foo()" } !scope1 = metadata !{ metadata !"scope 1", metadata !scope0 } !scope2 = metadata !{ metadata !"scope 2", metadata !scope0 } !scope3 = metadata !{ metadata !"scope 2.1", metadata !scope2 } !scope4 = metadata !{ metadata !"scope 2.2", metadata !scope2 } Loads and stores can be tagged with an alias-analysis scope, and also, with a noalias tag for a specific scope: ... = load %ptr1, !alias.scope !{ !scope1 } ... = load %ptr2, !alias.scope !{ !scope1, !scope2 }, !noalias !{ !scope1 } When evaluating an aliasing query, if one of the instructions is associated with an alias.scope id that is identical to the noalias scope associated with the other instruction, or is a descendant (in the scope hierarchy) of the noalias scope associated with the other instruction, then the two memory accesses are assumed not to alias. Note that is the first element of the scope metadata is a string, then it can be combined accross functions and translation units. The string can be replaced by a self-reference to create globally unqiue scope identifiers. [Note: This overview is slightly stylized, since the metadata nodes really need to just be numbers (!0 instead of !scope0), and the scope lists are also global unnamed metadata.] Existing noalias metadata in a callee is "cloned" for use by the inlined code. This is necessary because the aliasing scopes are unique to each call site (because of possible control dependencies on the aliasing properties). For example, consider a function: foo(noalias a, noalias b) { *a = *b; } that gets inlined into bar() { ... if (...) foo(a1, b1); ... if (...) foo(a2, b2); } -- now just because we know that a1 does not alias with b1 at the first call site, and a2 does not alias with b2 at the second call site, we cannot let inlining these functons have the metadata imply that a1 does not alias with b2. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@213864 91177308-0d34-0410-b5e6-96231b3b80d8
2014-07-24 14:25:39 +00:00
static MDNode *concatenate(MDNode *A, MDNode *B);
static MDNode *intersect(MDNode *A, MDNode *B);
static MDNode *getMostGenericTBAA(MDNode *A, MDNode *B);
static MDNode *getMostGenericFPMath(MDNode *A, MDNode *B);
static MDNode *getMostGenericRange(MDNode *A, MDNode *B);
};
template <class NodeTy>
typename std::enable_if<std::is_base_of<UniquableMDNode, NodeTy>::value,
NodeTy *>::type
MDNode::replaceWithUniqued(std::unique_ptr<NodeTy, TempMDNodeDeleter> Node) {
// Try to uniquify in place.
UniquableMDNode *UniquedNode = Node->uniquify();
if (UniquedNode == Node.get()) {
Node->makeUniqued();
return Node.release();
}
// Collision, so RAUW instead.
Node->replaceAllUsesWith(UniquedNode);
return cast<NodeTy>(UniquedNode);
}
template <class NodeTy>
typename std::enable_if<std::is_base_of<UniquableMDNode, NodeTy>::value,
NodeTy *>::type
MDNode::replaceWithDistinct(std::unique_ptr<NodeTy, TempMDNodeDeleter> Node) {
Node->makeDistinct();
return Node.release();
}
/// \brief Uniquable metadata node.
///
/// A uniquable metadata node. This contains the basic functionality
/// for implementing sub-types of \a MDNode that can be uniqued like
/// constants.
///
/// There is limited support for RAUW at construction time. At construction
/// time, if any operand is a temporary node (or an unresolved uniqued node,
/// which indicates a transitive temporary operand), the node itself will be
/// unresolved. As soon as all operands become resolved, it will drop RAUW
/// support permanently.
///
/// If an unresolved node is part of a cycle, \a resolveCycles() needs
/// to be called on some member of the cycle once all temporary nodes have been
/// replaced.
class UniquableMDNode : public MDNode {
friend class ReplaceableMetadataImpl;
friend class MDNode;
friend class LLVMContextImpl;
protected:
/// \brief Create a new node.
///
/// If \c AllowRAUW, then if any operands are unresolved support RAUW. RAUW
/// will be dropped once all operands have been resolved (or if \a
/// resolveCycles() is called).
UniquableMDNode(LLVMContext &C, unsigned ID, StorageType Storage,
ArrayRef<Metadata *> Vals);
~UniquableMDNode() {}
void storeDistinctInContext();
template <class T, class StoreT>
static T *storeImpl(T *N, StorageType Storage, StoreT &Store);
public:
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
static bool classof(const Metadata *MD) {
return MD->getMetadataID() == MDTupleKind ||
MD->getMetadataID() == MDLocationKind;
}
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
/// \brief Resolve cycles.
///
/// Once all forward declarations have been resolved, force cycles to be
/// resolved.
///
/// \pre No operands (or operands' operands, etc.) have \a isTemporary().
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
void resolveCycles();
private:
void handleChangedOperand(void *Ref, Metadata *New);
void resolve();
void resolveAfterOperandChange(Metadata *Old, Metadata *New);
void decrementUnresolvedOperandCount();
unsigned countUnresolvedOperands() const;
/// \brief Mutate this to be "uniqued".
///
/// Mutate this so that \a isUniqued().
/// \pre \a isTemporary().
/// \pre already added to uniquing set.
void makeUniqued();
/// \brief Mutate this to be "distinct".
///
/// Mutate this so that \a isDistinct().
/// \pre \a isTemporary().
void makeDistinct();
void deleteAsSubclass();
UniquableMDNode *uniquify();
void eraseFromStore();
};
/// \brief Tuple of metadata.
///
/// This is the simple \a MDNode arbitrary tuple. Nodes are uniqued by
/// default based on their operands.
class MDTuple : public UniquableMDNode {
friend class LLVMContextImpl;
friend class UniquableMDNode;
MDTuple(LLVMContext &C, StorageType Storage, unsigned Hash,
ArrayRef<Metadata *> Vals)
: UniquableMDNode(C, MDTupleKind, Storage, Vals) {
setHash(Hash);
}
~MDTuple() { dropAllReferences(); }
void setHash(unsigned Hash) { MDNodeSubclassData = Hash; }
void recalculateHash();
static MDTuple *getImpl(LLVMContext &Context, ArrayRef<Metadata *> MDs,
StorageType Storage, bool ShouldCreate = true);
public:
/// \brief Get the hash, if any.
unsigned getHash() const { return MDNodeSubclassData; }
static MDTuple *get(LLVMContext &Context, ArrayRef<Metadata *> MDs) {
return getImpl(Context, MDs, Uniqued);
}
static MDTuple *getIfExists(LLVMContext &Context, ArrayRef<Metadata *> MDs) {
return getImpl(Context, MDs, Uniqued, /* ShouldCreate */ false);
}
/// \brief Return a distinct node.
///
/// Return a distinct node -- i.e., a node that is not uniqued.
static MDTuple *getDistinct(LLVMContext &Context, ArrayRef<Metadata *> MDs) {
return getImpl(Context, MDs, Distinct);
}
/// \brief Return a temporary node.
///
/// For use in constructing cyclic MDNode structures. A temporary MDNode is
/// not uniqued, may be RAUW'd, and must be manually deleted with
/// deleteTemporary.
static TempMDTuple getTemporary(LLVMContext &Context,
ArrayRef<Metadata *> MDs) {
return TempMDTuple(getImpl(Context, MDs, Temporary));
}
static bool classof(const Metadata *MD) {
return MD->getMetadataID() == MDTupleKind;
}
private:
MDTuple *uniquifyImpl();
void eraseFromStoreImpl();
};
MDTuple *MDNode::get(LLVMContext &Context, ArrayRef<Metadata *> MDs) {
return MDTuple::get(Context, MDs);
}
MDTuple *MDNode::getIfExists(LLVMContext &Context, ArrayRef<Metadata *> MDs) {
return MDTuple::getIfExists(Context, MDs);
}
MDTuple *MDNode::getDistinct(LLVMContext &Context, ArrayRef<Metadata *> MDs) {
return MDTuple::getDistinct(Context, MDs);
}
TempMDTuple MDNode::getTemporary(LLVMContext &Context,
ArrayRef<Metadata *> MDs) {
return MDTuple::getTemporary(Context, MDs);
}
void TempMDNodeDeleter::operator()(MDNode *Node) const {
MDNode::deleteTemporary(Node);
}
/// \brief Debug location.
///
/// A debug location in source code, used for debug info and otherwise.
class MDLocation : public UniquableMDNode {
friend class LLVMContextImpl;
friend class UniquableMDNode;
MDLocation(LLVMContext &C, StorageType Storage, unsigned Line,
unsigned Column, ArrayRef<Metadata *> MDs);
~MDLocation() { dropAllReferences(); }
static MDLocation *getImpl(LLVMContext &Context, unsigned Line,
unsigned Column, Metadata *Scope,
Metadata *InlinedAt, StorageType Storage,
bool ShouldCreate = true);
// Disallow replacing operands.
void replaceOperandWith(unsigned I, Metadata *New) LLVM_DELETED_FUNCTION;
public:
static MDLocation *get(LLVMContext &Context, unsigned Line, unsigned Column,
Metadata *Scope, Metadata *InlinedAt = nullptr) {
return getImpl(Context, Line, Column, Scope, InlinedAt, Uniqued);
}
static MDLocation *getIfExists(LLVMContext &Context, unsigned Line,
unsigned Column, Metadata *Scope,
Metadata *InlinedAt = nullptr) {
return getImpl(Context, Line, Column, Scope, InlinedAt, Uniqued,
/* ShouldCreate */ false);
}
static MDLocation *getDistinct(LLVMContext &Context, unsigned Line,
unsigned Column, Metadata *Scope,
Metadata *InlinedAt = nullptr) {
return getImpl(Context, Line, Column, Scope, InlinedAt, Distinct);
}
static TempMDLocation getTemporary(LLVMContext &Context, unsigned Line,
unsigned Column, Metadata *Scope,
Metadata *InlinedAt = nullptr) {
return TempMDLocation(
getImpl(Context, Line, Column, Scope, InlinedAt, Temporary));
}
unsigned getLine() const { return MDNodeSubclassData; }
unsigned getColumn() const { return SubclassData16; }
Metadata *getScope() const { return getOperand(0); }
Metadata *getInlinedAt() const {
if (getNumOperands() == 2)
return getOperand(1);
return nullptr;
}
static bool classof(const Metadata *MD) {
return MD->getMetadataID() == MDLocationKind;
}
private:
MDLocation *uniquifyImpl();
void eraseFromStoreImpl();
};
//===----------------------------------------------------------------------===//
/// \brief A tuple of MDNodes.
///
/// Despite its name, a NamedMDNode isn't itself an MDNode. NamedMDNodes belong
/// to modules, have names, and contain lists of MDNodes.
///
/// TODO: Inherit from Metadata.
class NamedMDNode : public ilist_node<NamedMDNode> {
friend class SymbolTableListTraits<NamedMDNode, Module>;
friend struct ilist_traits<NamedMDNode>;
friend class LLVMContextImpl;
friend class Module;
NamedMDNode(const NamedMDNode &) LLVM_DELETED_FUNCTION;
std::string Name;
Module *Parent;
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.) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00
void *Operands; // SmallVector<TrackingMDRef, 4>
void setParent(Module *M) { Parent = M; }
explicit NamedMDNode(const Twine &N);
template<class T1, class T2>
class op_iterator_impl :
public std::iterator<std::bidirectional_iterator_tag, T2> {
const NamedMDNode *Node;
unsigned Idx;
op_iterator_impl(const NamedMDNode *N, unsigned i) : Node(N), Idx(i) { }
friend class NamedMDNode;
public:
op_iterator_impl() : Node(nullptr), Idx(0) { }
bool operator==(const op_iterator_impl &o) const { return Idx == o.Idx; }
bool operator!=(const op_iterator_impl &o) const { return Idx != o.Idx; }
op_iterator_impl &operator++() {
++Idx;
return *this;
}
op_iterator_impl operator++(int) {
op_iterator_impl tmp(*this);
operator++();
return tmp;
}
op_iterator_impl &operator--() {
--Idx;
return *this;
}
op_iterator_impl operator--(int) {
op_iterator_impl tmp(*this);
operator--();
return tmp;
}
T1 operator*() const { return Node->getOperand(Idx); }
};
public:
/// \brief Drop all references and remove the node from parent module.
void eraseFromParent();
/// \brief Remove all uses and clear node vector.
void dropAllReferences();
~NamedMDNode();
/// \brief Get the module that holds this named metadata collection.
inline Module *getParent() { return Parent; }
inline const Module *getParent() const { return Parent; }
MDNode *getOperand(unsigned i) const;
unsigned getNumOperands() const;
void addOperand(MDNode *M);
void setOperand(unsigned I, MDNode *New);
StringRef getName() const;
void print(raw_ostream &ROS) const;
void dump() const;
// ---------------------------------------------------------------------------
// Operand Iterator interface...
//
typedef op_iterator_impl<MDNode *, MDNode> op_iterator;
op_iterator op_begin() { return op_iterator(this, 0); }
op_iterator op_end() { return op_iterator(this, getNumOperands()); }
typedef op_iterator_impl<const MDNode *, MDNode> const_op_iterator;
const_op_iterator op_begin() const { return const_op_iterator(this, 0); }
const_op_iterator op_end() const { return const_op_iterator(this, getNumOperands()); }
inline iterator_range<op_iterator> operands() {
return iterator_range<op_iterator>(op_begin(), op_end());
}
inline iterator_range<const_op_iterator> operands() const {
return iterator_range<const_op_iterator>(op_begin(), op_end());
}
};
} // end llvm namespace
#endif