1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-19 02:52:53 +02:00
llvm-mirror/unittests/IR/MetadataTest.cpp

631 lines
19 KiB
C++
Raw Normal View History

//===- llvm/unittest/IR/Metadata.cpp - Metadata unit tests ----------------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
#include "llvm/ADT/STLExtras.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/LLVMContext.h"
#include "llvm/IR/Metadata.h"
#include "llvm/IR/Module.h"
#include "llvm/IR/Type.h"
#include "llvm/Support/raw_ostream.h"
#include "gtest/gtest.h"
using namespace llvm;
namespace {
TEST(ContextAndReplaceableUsesTest, FromContext) {
LLVMContext Context;
ContextAndReplaceableUses CRU(Context);
EXPECT_EQ(&Context, &CRU.getContext());
EXPECT_FALSE(CRU.hasReplaceableUses());
EXPECT_FALSE(CRU.getReplaceableUses());
}
TEST(ContextAndReplaceableUsesTest, FromReplaceableUses) {
LLVMContext Context;
ContextAndReplaceableUses CRU(make_unique<ReplaceableMetadataImpl>(Context));
EXPECT_EQ(&Context, &CRU.getContext());
EXPECT_TRUE(CRU.hasReplaceableUses());
EXPECT_TRUE(CRU.getReplaceableUses());
}
TEST(ContextAndReplaceableUsesTest, makeReplaceable) {
LLVMContext Context;
ContextAndReplaceableUses CRU(Context);
CRU.makeReplaceable(make_unique<ReplaceableMetadataImpl>(Context));
EXPECT_EQ(&Context, &CRU.getContext());
EXPECT_TRUE(CRU.hasReplaceableUses());
EXPECT_TRUE(CRU.getReplaceableUses());
}
TEST(ContextAndReplaceableUsesTest, takeReplaceableUses) {
LLVMContext Context;
auto ReplaceableUses = make_unique<ReplaceableMetadataImpl>(Context);
auto *Ptr = ReplaceableUses.get();
ContextAndReplaceableUses CRU(std::move(ReplaceableUses));
ReplaceableUses = CRU.takeReplaceableUses();
EXPECT_EQ(&Context, &CRU.getContext());
EXPECT_FALSE(CRU.hasReplaceableUses());
EXPECT_FALSE(CRU.getReplaceableUses());
EXPECT_EQ(Ptr, ReplaceableUses.get());
}
class MetadataTest : public testing::Test {
protected:
LLVMContext Context;
IR: Make MDNode::dump() useful by adding addresses It's horrible to inspect `MDNode`s in a debugger. All of their operands that are `MDNode`s get dumped as `<badref>`, since we can't assign metadata slots in the context of a `Metadata::dump()`. (Why not? Why not assign numbers lazily? Because then each time you called `dump()`, a given `MDNode` could have a different lazily assigned number.) Fortunately, the C memory model gives us perfectly good identifiers for `MDNode`. Add pointer addresses to the dumps, transforming this: (lldb) e N->dump() !{i32 662302, i32 26, <badref>, null} (lldb) e ((MDNode*)N->getOperand(2))->dump() !{i32 4, !"foo"} into: (lldb) e N->dump() !{i32 662302, i32 26, <0x100706ee0>, null} (lldb) e ((MDNode*)0x100706ee0)->dump() !{i32 4, !"foo"} and this: (lldb) e N->dump() 0x101200248 = !{<badref>, <badref>, <badref>, <badref>, <badref>} (lldb) e N->getOperand(0) (const llvm::MDOperand) $0 = { MD = 0x00000001012004e0 } (lldb) e N->getOperand(1) (const llvm::MDOperand) $1 = { MD = 0x00000001012004e0 } (lldb) e N->getOperand(2) (const llvm::MDOperand) $2 = { MD = 0x0000000101200058 } (lldb) e N->getOperand(3) (const llvm::MDOperand) $3 = { MD = 0x00000001012004e0 } (lldb) e N->getOperand(4) (const llvm::MDOperand) $4 = { MD = 0x0000000101200058 } (lldb) e ((MDNode*)0x00000001012004e0)->dump() !{} (lldb) e ((MDNode*)0x0000000101200058)->dump() !{null} into: (lldb) e N->dump() !{<0x1012004e0>, <0x1012004e0>, <0x101200058>, <0x1012004e0>, <0x101200058>} (lldb) e ((MDNode*)0x1012004e0)->dump() !{} (lldb) e ((MDNode*)0x101200058)->dump() !{null} llvm-svn: 224325
2014-12-16 08:09:37 +01:00
MDNode *getNode() { return MDNode::get(Context, None); }
MDNode *getNode(Metadata *MD) { return MDNode::get(Context, MD); }
MDNode *getNode(Metadata *MD1, Metadata *MD2) {
Metadata *MDs[] = {MD1, MD2};
return MDNode::get(Context, MDs);
}
};
typedef MetadataTest MDStringTest;
2009-07-31 23:38:10 +02:00
// Test that construction of MDString with different value produces different
// MDString objects, even with the same string pointer and nulls in the string.
TEST_F(MDStringTest, CreateDifferent) {
char x[3] = { 'f', 0, 'A' };
2009-07-31 23:38:10 +02:00
MDString *s1 = MDString::get(Context, StringRef(&x[0], 3));
x[2] = 'B';
2009-07-31 23:38:10 +02:00
MDString *s2 = MDString::get(Context, StringRef(&x[0], 3));
EXPECT_NE(s1, s2);
}
// Test that creation of MDStrings with the same string contents produces the
// same MDString object, even with different pointers.
TEST_F(MDStringTest, CreateSame) {
char x[4] = { 'a', 'b', 'c', 'X' };
char y[4] = { 'a', 'b', 'c', 'Y' };
2009-07-31 23:38:10 +02:00
MDString *s1 = MDString::get(Context, StringRef(&x[0], 3));
MDString *s2 = MDString::get(Context, StringRef(&y[0], 3));
EXPECT_EQ(s1, s2);
}
// Test that MDString prints out the string we fed it.
TEST_F(MDStringTest, PrintingSimple) {
char *str = new char[13];
strncpy(str, "testing 1 2 3", 13);
2009-07-31 23:38:10 +02:00
MDString *s = MDString::get(Context, StringRef(str, 13));
strncpy(str, "aaaaaaaaaaaaa", 13);
delete[] str;
std::string Str;
raw_string_ostream oss(Str);
s->print(oss);
EXPECT_STREQ("!\"testing 1 2 3\"", oss.str().c_str());
}
// Test printing of MDString with non-printable characters.
TEST_F(MDStringTest, PrintingComplex) {
char str[5] = {0, '\n', '"', '\\', (char)-1};
2009-07-31 23:38:10 +02:00
MDString *s = MDString::get(Context, StringRef(str+0, 5));
std::string Str;
raw_string_ostream oss(Str);
s->print(oss);
EXPECT_STREQ("!\"\\00\\0A\\22\\5C\\FF\"", oss.str().c_str());
}
typedef MetadataTest MDNodeTest;
// Test the two constructors, and containing other Constants.
TEST_F(MDNodeTest, Simple) {
char x[3] = { 'a', 'b', 'c' };
char y[3] = { '1', '2', '3' };
2009-07-31 23:38:10 +02:00
MDString *s1 = MDString::get(Context, StringRef(&x[0], 3));
MDString *s2 = MDString::get(Context, StringRef(&y[0], 3));
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
ConstantAsMetadata *CI = ConstantAsMetadata::get(
ConstantInt::get(getGlobalContext(), APInt(8, 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.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
std::vector<Metadata *> V;
V.push_back(s1);
V.push_back(CI);
V.push_back(s2);
MDNode *n1 = MDNode::get(Context, V);
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
Metadata *const c1 = n1;
MDNode *n2 = MDNode::get(Context, c1);
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
Metadata *const c2 = n2;
MDNode *n3 = MDNode::get(Context, V);
MDNode *n4 = MDNode::getIfExists(Context, V);
MDNode *n5 = MDNode::getIfExists(Context, c1);
MDNode *n6 = MDNode::getIfExists(Context, c2);
EXPECT_NE(n1, n2);
EXPECT_EQ(n1, n3);
EXPECT_EQ(n4, n1);
EXPECT_EQ(n5, n2);
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
EXPECT_EQ(n6, (Metadata *)nullptr);
EXPECT_EQ(3u, n1->getNumOperands());
EXPECT_EQ(s1, n1->getOperand(0));
EXPECT_EQ(CI, n1->getOperand(1));
EXPECT_EQ(s2, n1->getOperand(2));
EXPECT_EQ(1u, n2->getNumOperands());
EXPECT_EQ(n1, n2->getOperand(0));
}
TEST_F(MDNodeTest, Delete) {
Constant *C = ConstantInt::get(Type::getInt32Ty(getGlobalContext()), 1);
Instruction *I = new BitCastInst(C, Type::getInt32Ty(getGlobalContext()));
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
Metadata *const V = LocalAsMetadata::get(I);
MDNode *n = MDNode::get(Context, V);
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
TrackingMDRef wvh(n);
EXPECT_EQ(n, wvh);
delete I;
}
2009-07-30 02:03:41 +02:00
TEST_F(MDNodeTest, SelfReference) {
// !0 = !{!0}
// !1 = !{!0}
{
MDNode *Temp = MDNode::getTemporary(Context, None);
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
Metadata *Args[] = {Temp};
MDNode *Self = MDNode::get(Context, Args);
Self->replaceOperandWith(0, Self);
MDNode::deleteTemporary(Temp);
ASSERT_EQ(Self, Self->getOperand(0));
// Self-references should be distinct, so MDNode::get() should grab a
// uniqued node that references Self, not Self.
Args[0] = Self;
MDNode *Ref1 = MDNode::get(Context, Args);
MDNode *Ref2 = MDNode::get(Context, Args);
EXPECT_NE(Self, Ref1);
EXPECT_EQ(Ref1, Ref2);
}
// !0 = !{!0, !{}}
// !1 = !{!0, !{}}
{
MDNode *Temp = MDNode::getTemporary(Context, None);
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
Metadata *Args[] = {Temp, MDNode::get(Context, None)};
MDNode *Self = MDNode::get(Context, Args);
Self->replaceOperandWith(0, Self);
MDNode::deleteTemporary(Temp);
ASSERT_EQ(Self, Self->getOperand(0));
// Self-references should be distinct, so MDNode::get() should grab a
// uniqued node that references Self, not Self itself.
Args[0] = Self;
MDNode *Ref1 = MDNode::get(Context, Args);
MDNode *Ref2 = MDNode::get(Context, Args);
EXPECT_NE(Self, Ref1);
EXPECT_EQ(Ref1, Ref2);
}
}
IR: Make MDNode::dump() useful by adding addresses It's horrible to inspect `MDNode`s in a debugger. All of their operands that are `MDNode`s get dumped as `<badref>`, since we can't assign metadata slots in the context of a `Metadata::dump()`. (Why not? Why not assign numbers lazily? Because then each time you called `dump()`, a given `MDNode` could have a different lazily assigned number.) Fortunately, the C memory model gives us perfectly good identifiers for `MDNode`. Add pointer addresses to the dumps, transforming this: (lldb) e N->dump() !{i32 662302, i32 26, <badref>, null} (lldb) e ((MDNode*)N->getOperand(2))->dump() !{i32 4, !"foo"} into: (lldb) e N->dump() !{i32 662302, i32 26, <0x100706ee0>, null} (lldb) e ((MDNode*)0x100706ee0)->dump() !{i32 4, !"foo"} and this: (lldb) e N->dump() 0x101200248 = !{<badref>, <badref>, <badref>, <badref>, <badref>} (lldb) e N->getOperand(0) (const llvm::MDOperand) $0 = { MD = 0x00000001012004e0 } (lldb) e N->getOperand(1) (const llvm::MDOperand) $1 = { MD = 0x00000001012004e0 } (lldb) e N->getOperand(2) (const llvm::MDOperand) $2 = { MD = 0x0000000101200058 } (lldb) e N->getOperand(3) (const llvm::MDOperand) $3 = { MD = 0x00000001012004e0 } (lldb) e N->getOperand(4) (const llvm::MDOperand) $4 = { MD = 0x0000000101200058 } (lldb) e ((MDNode*)0x00000001012004e0)->dump() !{} (lldb) e ((MDNode*)0x0000000101200058)->dump() !{null} into: (lldb) e N->dump() !{<0x1012004e0>, <0x1012004e0>, <0x101200058>, <0x1012004e0>, <0x101200058>} (lldb) e ((MDNode*)0x1012004e0)->dump() !{} (lldb) e ((MDNode*)0x101200058)->dump() !{null} llvm-svn: 224325
2014-12-16 08:09:37 +01:00
TEST_F(MDNodeTest, Print) {
Constant *C = ConstantInt::get(Type::getInt32Ty(Context), 7);
MDString *S = MDString::get(Context, "foo");
MDNode *N0 = getNode();
MDNode *N1 = getNode(N0);
MDNode *N2 = getNode(N0, N1);
Metadata *Args[] = {ConstantAsMetadata::get(C), S, nullptr, N0, N1, N2};
MDNode *N = MDNode::get(Context, Args);
std::string Expected;
{
raw_string_ostream OS(Expected);
OS << "!{";
IR: Make MDNode::dump() useful by adding addresses It's horrible to inspect `MDNode`s in a debugger. All of their operands that are `MDNode`s get dumped as `<badref>`, since we can't assign metadata slots in the context of a `Metadata::dump()`. (Why not? Why not assign numbers lazily? Because then each time you called `dump()`, a given `MDNode` could have a different lazily assigned number.) Fortunately, the C memory model gives us perfectly good identifiers for `MDNode`. Add pointer addresses to the dumps, transforming this: (lldb) e N->dump() !{i32 662302, i32 26, <badref>, null} (lldb) e ((MDNode*)N->getOperand(2))->dump() !{i32 4, !"foo"} into: (lldb) e N->dump() !{i32 662302, i32 26, <0x100706ee0>, null} (lldb) e ((MDNode*)0x100706ee0)->dump() !{i32 4, !"foo"} and this: (lldb) e N->dump() 0x101200248 = !{<badref>, <badref>, <badref>, <badref>, <badref>} (lldb) e N->getOperand(0) (const llvm::MDOperand) $0 = { MD = 0x00000001012004e0 } (lldb) e N->getOperand(1) (const llvm::MDOperand) $1 = { MD = 0x00000001012004e0 } (lldb) e N->getOperand(2) (const llvm::MDOperand) $2 = { MD = 0x0000000101200058 } (lldb) e N->getOperand(3) (const llvm::MDOperand) $3 = { MD = 0x00000001012004e0 } (lldb) e N->getOperand(4) (const llvm::MDOperand) $4 = { MD = 0x0000000101200058 } (lldb) e ((MDNode*)0x00000001012004e0)->dump() !{} (lldb) e ((MDNode*)0x0000000101200058)->dump() !{null} into: (lldb) e N->dump() !{<0x1012004e0>, <0x1012004e0>, <0x101200058>, <0x1012004e0>, <0x101200058>} (lldb) e ((MDNode*)0x1012004e0)->dump() !{} (lldb) e ((MDNode*)0x101200058)->dump() !{null} llvm-svn: 224325
2014-12-16 08:09:37 +01:00
C->printAsOperand(OS);
OS << ", ";
S->printAsOperand(OS);
IR: Make MDNode::dump() useful by adding addresses It's horrible to inspect `MDNode`s in a debugger. All of their operands that are `MDNode`s get dumped as `<badref>`, since we can't assign metadata slots in the context of a `Metadata::dump()`. (Why not? Why not assign numbers lazily? Because then each time you called `dump()`, a given `MDNode` could have a different lazily assigned number.) Fortunately, the C memory model gives us perfectly good identifiers for `MDNode`. Add pointer addresses to the dumps, transforming this: (lldb) e N->dump() !{i32 662302, i32 26, <badref>, null} (lldb) e ((MDNode*)N->getOperand(2))->dump() !{i32 4, !"foo"} into: (lldb) e N->dump() !{i32 662302, i32 26, <0x100706ee0>, null} (lldb) e ((MDNode*)0x100706ee0)->dump() !{i32 4, !"foo"} and this: (lldb) e N->dump() 0x101200248 = !{<badref>, <badref>, <badref>, <badref>, <badref>} (lldb) e N->getOperand(0) (const llvm::MDOperand) $0 = { MD = 0x00000001012004e0 } (lldb) e N->getOperand(1) (const llvm::MDOperand) $1 = { MD = 0x00000001012004e0 } (lldb) e N->getOperand(2) (const llvm::MDOperand) $2 = { MD = 0x0000000101200058 } (lldb) e N->getOperand(3) (const llvm::MDOperand) $3 = { MD = 0x00000001012004e0 } (lldb) e N->getOperand(4) (const llvm::MDOperand) $4 = { MD = 0x0000000101200058 } (lldb) e ((MDNode*)0x00000001012004e0)->dump() !{} (lldb) e ((MDNode*)0x0000000101200058)->dump() !{null} into: (lldb) e N->dump() !{<0x1012004e0>, <0x1012004e0>, <0x101200058>, <0x1012004e0>, <0x101200058>} (lldb) e ((MDNode*)0x1012004e0)->dump() !{} (lldb) e ((MDNode*)0x101200058)->dump() !{null} llvm-svn: 224325
2014-12-16 08:09:37 +01:00
OS << ", null";
MDNode *Nodes[] = {N0, N1, N2};
for (auto *Node : Nodes)
OS << ", <" << (void *)Node << ">";
OS << "}\n";
}
std::string Actual;
{
raw_string_ostream OS(Actual);
N->print(OS);
}
EXPECT_EQ(Expected, Actual);
}
TEST_F(MDNodeTest, NullOperand) {
// metadata !{}
MDNode *Empty = MDNode::get(Context, None);
// metadata !{metadata !{}}
Metadata *Ops[] = {Empty};
MDNode *N = MDNode::get(Context, Ops);
ASSERT_EQ(Empty, N->getOperand(0));
// metadata !{metadata !{}} => metadata !{null}
N->replaceOperandWith(0, nullptr);
ASSERT_EQ(nullptr, N->getOperand(0));
// metadata !{null}
Ops[0] = nullptr;
MDNode *NullOp = MDNode::get(Context, Ops);
ASSERT_EQ(nullptr, NullOp->getOperand(0));
EXPECT_EQ(N, NullOp);
}
TEST_F(MDNodeTest, DistinctOnUniquingCollision) {
// !{}
MDNode *Empty = MDNode::get(Context, None);
ASSERT_TRUE(Empty->isResolved());
EXPECT_FALSE(Empty->isDistinct());
// !{!{}}
Metadata *Wrapped1Ops[] = {Empty};
MDNode *Wrapped1 = MDNode::get(Context, Wrapped1Ops);
ASSERT_EQ(Empty, Wrapped1->getOperand(0));
ASSERT_TRUE(Wrapped1->isResolved());
EXPECT_FALSE(Wrapped1->isDistinct());
// !{!{!{}}}
Metadata *Wrapped2Ops[] = {Wrapped1};
MDNode *Wrapped2 = MDNode::get(Context, Wrapped2Ops);
ASSERT_EQ(Wrapped1, Wrapped2->getOperand(0));
ASSERT_TRUE(Wrapped2->isResolved());
EXPECT_FALSE(Wrapped2->isDistinct());
// !{!{!{}}} => !{!{}}
Wrapped2->replaceOperandWith(0, Empty);
ASSERT_EQ(Empty, Wrapped2->getOperand(0));
EXPECT_TRUE(Wrapped2->isDistinct());
EXPECT_FALSE(Wrapped1->isDistinct());
}
TEST_F(MDNodeTest, getDistinct) {
// !{}
MDNode *Empty = MDNode::get(Context, None);
ASSERT_TRUE(Empty->isResolved());
ASSERT_FALSE(Empty->isDistinct());
ASSERT_EQ(Empty, MDNode::get(Context, None));
// distinct !{}
MDNode *Distinct1 = MDNode::getDistinct(Context, None);
MDNode *Distinct2 = MDNode::getDistinct(Context, None);
EXPECT_TRUE(Distinct1->isResolved());
EXPECT_TRUE(Distinct2->isDistinct());
EXPECT_NE(Empty, Distinct1);
EXPECT_NE(Empty, Distinct2);
EXPECT_NE(Distinct1, Distinct2);
// !{}
ASSERT_EQ(Empty, MDNode::get(Context, None));
}
TEST_F(MDNodeTest, isUniqued) {
MDNode *U = MDTuple::get(Context, None);
MDNode *D = MDTuple::getDistinct(Context, None);
MDNode *T = MDTuple::getTemporary(Context, None);
EXPECT_TRUE(U->isUniqued());
EXPECT_FALSE(D->isUniqued());
EXPECT_FALSE(T->isUniqued());
MDNode::deleteTemporary(T);
}
TEST_F(MDNodeTest, isDistinct) {
MDNode *U = MDTuple::get(Context, None);
MDNode *D = MDTuple::getDistinct(Context, None);
MDNode *T = MDTuple::getTemporary(Context, None);
EXPECT_FALSE(U->isDistinct());
EXPECT_TRUE(D->isDistinct());
EXPECT_FALSE(T->isDistinct());
MDNode::deleteTemporary(T);
}
TEST_F(MDNodeTest, isTemporary) {
MDNode *U = MDTuple::get(Context, None);
MDNode *D = MDTuple::getDistinct(Context, None);
MDNode *T = MDTuple::getTemporary(Context, None);
EXPECT_FALSE(U->isTemporary());
EXPECT_FALSE(D->isTemporary());
EXPECT_TRUE(T->isTemporary());
MDNode::deleteTemporary(T);
}
TEST_F(MDNodeTest, getDistinctWithUnresolvedOperands) {
// temporary !{}
MDTuple *Temp = MDTuple::getTemporary(Context, None);
ASSERT_FALSE(Temp->isResolved());
// distinct !{temporary !{}}
Metadata *Ops[] = {Temp};
MDNode *Distinct = MDNode::getDistinct(Context, Ops);
EXPECT_TRUE(Distinct->isResolved());
EXPECT_EQ(Temp, Distinct->getOperand(0));
// temporary !{} => !{}
MDNode *Empty = MDNode::get(Context, None);
Temp->replaceAllUsesWith(Empty);
MDNode::deleteTemporary(Temp);
EXPECT_EQ(Empty, Distinct->getOperand(0));
}
TEST_F(MDNodeTest, handleChangedOperandRecursion) {
// !0 = !{}
MDNode *N0 = MDNode::get(Context, None);
// !1 = !{!3, null}
MDTuple *Temp3 = MDTuple::getTemporary(Context, None);
Metadata *Ops1[] = {Temp3, nullptr};
MDNode *N1 = MDNode::get(Context, Ops1);
// !2 = !{!3, !0}
Metadata *Ops2[] = {Temp3, N0};
MDNode *N2 = MDNode::get(Context, Ops2);
// !3 = !{!2}
Metadata *Ops3[] = {N2};
MDNode *N3 = MDNode::get(Context, Ops3);
Temp3->replaceAllUsesWith(N3);
MDNode::deleteTemporary(Temp3);
// !4 = !{!1}
Metadata *Ops4[] = {N1};
MDNode *N4 = MDNode::get(Context, Ops4);
// Confirm that the cycle prevented RAUW from getting dropped.
EXPECT_TRUE(N0->isResolved());
EXPECT_FALSE(N1->isResolved());
EXPECT_FALSE(N2->isResolved());
EXPECT_FALSE(N3->isResolved());
EXPECT_FALSE(N4->isResolved());
// Create a couple of distinct nodes to observe what's going on.
//
// !5 = distinct !{!2}
// !6 = distinct !{!3}
Metadata *Ops5[] = {N2};
MDNode *N5 = MDNode::getDistinct(Context, Ops5);
Metadata *Ops6[] = {N3};
MDNode *N6 = MDNode::getDistinct(Context, Ops6);
// Mutate !2 to look like !1, causing a uniquing collision (and an RAUW).
// This will ripple up, with !3 colliding with !4, and RAUWing. Since !2
// references !3, this can cause a re-entry of handleChangedOperand() when !3
// is not ready for it.
//
// !2->replaceOperandWith(1, nullptr)
// !2: !{!3, !0} => !{!3, null}
// !2->replaceAllUsesWith(!1)
// !3: !{!2] => !{!1}
// !3->replaceAllUsesWith(!4)
N2->replaceOperandWith(1, nullptr);
// If all has gone well, N2 and N3 will have been RAUW'ed and deleted from
// under us. Just check that the other nodes are sane.
//
// !1 = !{!4, null}
// !4 = !{!1}
// !5 = distinct !{!1}
// !6 = distinct !{!4}
EXPECT_EQ(N4, N1->getOperand(0));
EXPECT_EQ(N1, N4->getOperand(0));
EXPECT_EQ(N1, N5->getOperand(0));
EXPECT_EQ(N4, N6->getOperand(0));
}
TEST_F(MDNodeTest, replaceResolvedOperand) {
// Check code for replacing one resolved operand with another. If doing this
// directly (via replaceOperandWith()) becomes illegal, change the operand to
// a global value that gets RAUW'ed.
//
// Use a temporary node to keep N from being resolved.
MDTuple *Temp = MDTuple::getTemporary(Context, None);
Metadata *Ops[] = {nullptr, Temp};
MDNode *Empty = MDTuple::get(Context, ArrayRef<Metadata *>());
MDNode *N = MDTuple::get(Context, Ops);
EXPECT_EQ(nullptr, N->getOperand(0));
ASSERT_FALSE(N->isResolved());
// Check code for replacing resolved nodes.
N->replaceOperandWith(0, Empty);
EXPECT_EQ(Empty, N->getOperand(0));
// Check code for adding another unresolved operand.
N->replaceOperandWith(0, Temp);
EXPECT_EQ(Temp, N->getOperand(0));
// Remove the references to Temp; required for teardown.
Temp->replaceAllUsesWith(nullptr);
MDNode::deleteTemporary(Temp);
}
typedef MetadataTest MDLocationTest;
TEST_F(MDLocationTest, Overflow) {
MDNode *N = MDNode::get(Context, None);
{
MDLocation *L = MDLocation::get(Context, 2, 7, N);
EXPECT_EQ(2u, L->getLine());
EXPECT_EQ(7u, L->getColumn());
}
unsigned U24 = 1u << 24;
unsigned U16 = 1u << 16;
{
MDLocation *L = MDLocation::get(Context, U24 - 1, U16 - 1, N);
EXPECT_EQ(U24 - 1, L->getLine());
EXPECT_EQ(U16 - 1, L->getColumn());
}
{
MDLocation *L = MDLocation::get(Context, U24, U16, N);
EXPECT_EQ(0u, L->getLine());
EXPECT_EQ(0u, L->getColumn());
}
{
MDLocation *L = MDLocation::get(Context, U24 + 1, U16 + 1, N);
EXPECT_EQ(0u, L->getLine());
EXPECT_EQ(0u, L->getColumn());
}
}
TEST_F(MDLocationTest, getDistinct) {
MDNode *N = MDNode::get(Context, None);
MDLocation *L0 = MDLocation::getDistinct(Context, 2, 7, N);
EXPECT_TRUE(L0->isDistinct());
MDLocation *L1 = MDLocation::get(Context, 2, 7, N);
EXPECT_FALSE(L1->isDistinct());
EXPECT_EQ(L1, MDLocation::get(Context, 2, 7, N));
}
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
typedef MetadataTest MetadataAsValueTest;
TEST_F(MetadataAsValueTest, MDNode) {
MDNode *N = MDNode::get(Context, None);
auto *V = MetadataAsValue::get(Context, N);
EXPECT_TRUE(V->getType()->isMetadataTy());
EXPECT_EQ(N, V->getMetadata());
auto *V2 = MetadataAsValue::get(Context, N);
EXPECT_EQ(V, V2);
}
TEST_F(MetadataAsValueTest, MDNodeMDNode) {
MDNode *N = MDNode::get(Context, None);
Metadata *Ops[] = {N};
MDNode *N2 = MDNode::get(Context, Ops);
auto *V = MetadataAsValue::get(Context, N2);
EXPECT_TRUE(V->getType()->isMetadataTy());
EXPECT_EQ(N2, V->getMetadata());
auto *V2 = MetadataAsValue::get(Context, N2);
EXPECT_EQ(V, V2);
auto *V3 = MetadataAsValue::get(Context, N);
EXPECT_TRUE(V3->getType()->isMetadataTy());
EXPECT_NE(V, V3);
EXPECT_EQ(N, V3->getMetadata());
}
TEST_F(MetadataAsValueTest, MDNodeConstant) {
auto *C = ConstantInt::getTrue(Context);
auto *MD = ConstantAsMetadata::get(C);
Metadata *Ops[] = {MD};
auto *N = MDNode::get(Context, Ops);
auto *V = MetadataAsValue::get(Context, MD);
EXPECT_TRUE(V->getType()->isMetadataTy());
EXPECT_EQ(MD, V->getMetadata());
auto *V2 = MetadataAsValue::get(Context, N);
EXPECT_EQ(MD, V2->getMetadata());
EXPECT_EQ(V, V2);
}
typedef MetadataTest ValueAsMetadataTest;
TEST_F(ValueAsMetadataTest, UpdatesOnRAUW) {
Type *Ty = Type::getInt1PtrTy(Context);
std::unique_ptr<GlobalVariable> GV0(
new GlobalVariable(Ty, false, GlobalValue::ExternalLinkage));
auto *MD = ValueAsMetadata::get(GV0.get());
EXPECT_TRUE(MD->getValue() == GV0.get());
ASSERT_TRUE(GV0->use_empty());
std::unique_ptr<GlobalVariable> GV1(
new GlobalVariable(Ty, false, GlobalValue::ExternalLinkage));
GV0->replaceAllUsesWith(GV1.get());
EXPECT_TRUE(MD->getValue() == GV1.get());
}
TEST_F(ValueAsMetadataTest, CollidingDoubleUpdates) {
// Create a constant.
ConstantAsMetadata *CI = ConstantAsMetadata::get(
ConstantInt::get(getGlobalContext(), APInt(8, 0)));
// Create a temporary to prevent nodes from resolving.
MDTuple *Temp = MDTuple::getTemporary(Context, None);
// When the first operand of N1 gets reset to nullptr, it'll collide with N2.
Metadata *Ops1[] = {CI, CI, Temp};
Metadata *Ops2[] = {nullptr, CI, Temp};
auto *N1 = MDTuple::get(Context, Ops1);
auto *N2 = MDTuple::get(Context, Ops2);
ASSERT_NE(N1, N2);
// Tell metadata that the constant is getting deleted.
//
// After this, N1 will be invalid, so don't touch it.
ValueAsMetadata::handleDeletion(CI->getValue());
EXPECT_EQ(nullptr, N2->getOperand(0));
EXPECT_EQ(nullptr, N2->getOperand(1));
EXPECT_EQ(Temp, N2->getOperand(2));
// Clean up Temp for teardown.
Temp->replaceAllUsesWith(nullptr);
MDNode::deleteTemporary(Temp);
}
typedef MetadataTest TrackingMDRefTest;
TEST_F(TrackingMDRefTest, UpdatesOnRAUW) {
Type *Ty = Type::getInt1PtrTy(Context);
std::unique_ptr<GlobalVariable> GV0(
new GlobalVariable(Ty, false, GlobalValue::ExternalLinkage));
TypedTrackingMDRef<ValueAsMetadata> MD(ValueAsMetadata::get(GV0.get()));
EXPECT_TRUE(MD->getValue() == GV0.get());
ASSERT_TRUE(GV0->use_empty());
std::unique_ptr<GlobalVariable> GV1(
new GlobalVariable(Ty, false, GlobalValue::ExternalLinkage));
GV0->replaceAllUsesWith(GV1.get());
EXPECT_TRUE(MD->getValue() == GV1.get());
// Reset it, so we don't inadvertently test deletion.
MD.reset();
}
TEST_F(TrackingMDRefTest, UpdatesOnDeletion) {
Type *Ty = Type::getInt1PtrTy(Context);
std::unique_ptr<GlobalVariable> GV(
new GlobalVariable(Ty, false, GlobalValue::ExternalLinkage));
TypedTrackingMDRef<ValueAsMetadata> MD(ValueAsMetadata::get(GV.get()));
EXPECT_TRUE(MD->getValue() == GV.get());
ASSERT_TRUE(GV->use_empty());
GV.reset();
EXPECT_TRUE(!MD);
}
2009-07-30 02:03:41 +02:00
TEST(NamedMDNodeTest, Search) {
LLVMContext 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.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
ConstantAsMetadata *C =
ConstantAsMetadata::get(ConstantInt::get(Type::getInt32Ty(Context), 1));
ConstantAsMetadata *C2 =
ConstantAsMetadata::get(ConstantInt::get(Type::getInt32Ty(Context), 2));
2009-07-30 02:03:41 +02:00
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
Metadata *const V = C;
Metadata *const V2 = C2;
MDNode *n = MDNode::get(Context, V);
MDNode *n2 = MDNode::get(Context, V2);
2009-07-30 02:03:41 +02:00
Module M("MyModule", Context);
2009-07-30 02:03:41 +02:00
const char *Name = "llvm.NMD1";
NamedMDNode *NMD = M.getOrInsertNamedMetadata(Name);
NMD->addOperand(n);
NMD->addOperand(n2);
std::string Str;
raw_string_ostream oss(Str);
2009-07-30 02:03:41 +02:00
NMD->print(oss);
EXPECT_STREQ("!llvm.NMD1 = !{!0, !1}\n",
2009-07-30 02:03:41 +02:00
oss.str().c_str());
}
}