1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-25 04:02:41 +01:00
llvm-mirror/lib/IR/AsmWriter.cpp

3715 lines
122 KiB
C++
Raw Normal View History

//===- AsmWriter.cpp - Printing LLVM as an assembly file ------------------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
2001-06-06 22:29:01 +02:00
//
// This library implements `print` family of functions in classes like
// Module, Function, Value, etc. In-memory representation of those classes is
// converted to IR strings.
2001-06-06 22:29:01 +02:00
//
// Note that these routines must be extremely tolerant of various errors in the
// LLVM code, because it can be used for debugging transformations.
//
2001-06-06 22:29:01 +02:00
//===----------------------------------------------------------------------===//
#include "llvm/ADT/APFloat.h"
#include "llvm/ADT/APInt.h"
#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/None.h"
#include "llvm/ADT/Optional.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SetVector.h"
#include "llvm/ADT/SmallString.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/StringExtras.h"
#include "llvm/ADT/StringRef.h"
#include "llvm/ADT/iterator_range.h"
#include "llvm/BinaryFormat/Dwarf.h"
#include "llvm/Config/llvm-config.h"
#include "llvm/IR/Argument.h"
#include "llvm/IR/AssemblyAnnotationWriter.h"
#include "llvm/IR/Attributes.h"
#include "llvm/IR/BasicBlock.h"
#include "llvm/IR/CFG.h"
#include "llvm/IR/CallSite.h"
#include "llvm/IR/CallingConv.h"
#include "llvm/IR/Comdat.h"
#include "llvm/IR/Constant.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/DebugInfoMetadata.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/GlobalAlias.h"
#include "llvm/IR/GlobalIFunc.h"
#include "llvm/IR/GlobalIndirectSymbol.h"
#include "llvm/IR/GlobalObject.h"
#include "llvm/IR/GlobalValue.h"
#include "llvm/IR/GlobalVariable.h"
#include "llvm/IR/IRPrintingPasses.h"
#include "llvm/IR/InlineAsm.h"
#include "llvm/IR/InstrTypes.h"
#include "llvm/IR/Instruction.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/LLVMContext.h"
#include "llvm/IR/Metadata.h"
#include "llvm/IR/Module.h"
#include "llvm/IR/ModuleSlotTracker.h"
#include "llvm/IR/Operator.h"
#include "llvm/IR/Statepoint.h"
#include "llvm/IR/Type.h"
#include "llvm/IR/TypeFinder.h"
#include "llvm/IR/Use.h"
#include "llvm/IR/UseListOrder.h"
#include "llvm/IR/User.h"
#include "llvm/IR/Value.h"
#include "llvm/Support/AtomicOrdering.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/Format.h"
#include "llvm/Support/FormattedStream.h"
#include "llvm/Support/raw_ostream.h"
#include <algorithm>
#include <cassert>
#include <cctype>
#include <cstddef>
#include <cstdint>
#include <iterator>
#include <memory>
#include <string>
#include <tuple>
#include <utility>
#include <vector>
using namespace llvm;
// Make virtual table appear in this compilation unit.
AssemblyAnnotationWriter::~AssemblyAnnotationWriter() = default;
//===----------------------------------------------------------------------===//
// Helper Functions
//===----------------------------------------------------------------------===//
namespace {
struct OrderMap {
DenseMap<const Value *, std::pair<unsigned, bool>> IDs;
unsigned size() const { return IDs.size(); }
std::pair<unsigned, bool> &operator[](const Value *V) { return IDs[V]; }
std::pair<unsigned, bool> lookup(const Value *V) const {
return IDs.lookup(V);
}
void index(const Value *V) {
// Explicitly sequence get-size and insert-value operations to avoid UB.
unsigned ID = IDs.size() + 1;
IDs[V].first = ID;
}
};
} // end anonymous namespace
static void orderValue(const Value *V, OrderMap &OM) {
if (OM.lookup(V).first)
return;
if (const Constant *C = dyn_cast<Constant>(V))
if (C->getNumOperands() && !isa<GlobalValue>(C))
for (const Value *Op : C->operands())
if (!isa<BasicBlock>(Op) && !isa<GlobalValue>(Op))
orderValue(Op, OM);
// Note: we cannot cache this lookup above, since inserting into the map
// changes the map's size, and thus affects the other IDs.
OM.index(V);
}
static OrderMap orderModule(const Module *M) {
// This needs to match the order used by ValueEnumerator::ValueEnumerator()
// and ValueEnumerator::incorporateFunction().
OrderMap OM;
for (const GlobalVariable &G : M->globals()) {
if (G.hasInitializer())
if (!isa<GlobalValue>(G.getInitializer()))
orderValue(G.getInitializer(), OM);
orderValue(&G, OM);
}
for (const GlobalAlias &A : M->aliases()) {
if (!isa<GlobalValue>(A.getAliasee()))
orderValue(A.getAliasee(), OM);
orderValue(&A, OM);
}
for (const GlobalIFunc &I : M->ifuncs()) {
if (!isa<GlobalValue>(I.getResolver()))
orderValue(I.getResolver(), OM);
orderValue(&I, OM);
}
for (const Function &F : *M) {
for (const Use &U : F.operands())
if (!isa<GlobalValue>(U.get()))
orderValue(U.get(), OM);
orderValue(&F, OM);
if (F.isDeclaration())
continue;
for (const Argument &A : F.args())
orderValue(&A, OM);
for (const BasicBlock &BB : F) {
orderValue(&BB, OM);
for (const Instruction &I : BB) {
for (const Value *Op : I.operands())
if ((isa<Constant>(*Op) && !isa<GlobalValue>(*Op)) ||
isa<InlineAsm>(*Op))
orderValue(Op, OM);
orderValue(&I, OM);
}
}
}
return OM;
}
static void predictValueUseListOrderImpl(const Value *V, const Function *F,
unsigned ID, const OrderMap &OM,
UseListOrderStack &Stack) {
// Predict use-list order for this one.
using Entry = std::pair<const Use *, unsigned>;
SmallVector<Entry, 64> List;
for (const Use &U : V->uses())
// Check if this user will be serialized.
if (OM.lookup(U.getUser()).first)
List.push_back(std::make_pair(&U, List.size()));
if (List.size() < 2)
// We may have lost some users.
return;
bool GetsReversed =
!isa<GlobalVariable>(V) && !isa<Function>(V) && !isa<BasicBlock>(V);
if (auto *BA = dyn_cast<BlockAddress>(V))
ID = OM.lookup(BA->getBasicBlock()).first;
llvm::sort(List.begin(), List.end(), [&](const Entry &L, const Entry &R) {
const Use *LU = L.first;
const Use *RU = R.first;
if (LU == RU)
return false;
auto LID = OM.lookup(LU->getUser()).first;
auto RID = OM.lookup(RU->getUser()).first;
// If ID is 4, then expect: 7 6 5 1 2 3.
if (LID < RID) {
if (GetsReversed)
if (RID <= ID)
return true;
return false;
}
if (RID < LID) {
if (GetsReversed)
if (LID <= ID)
return false;
return true;
}
// LID and RID are equal, so we have different operands of the same user.
// Assume operands are added in order for all instructions.
if (GetsReversed)
if (LID <= ID)
return LU->getOperandNo() < RU->getOperandNo();
return LU->getOperandNo() > RU->getOperandNo();
});
if (std::is_sorted(
List.begin(), List.end(),
[](const Entry &L, const Entry &R) { return L.second < R.second; }))
// Order is already correct.
return;
// Store the shuffle.
Stack.emplace_back(V, F, List.size());
assert(List.size() == Stack.back().Shuffle.size() && "Wrong size");
for (size_t I = 0, E = List.size(); I != E; ++I)
Stack.back().Shuffle[I] = List[I].second;
}
static void predictValueUseListOrder(const Value *V, const Function *F,
OrderMap &OM, UseListOrderStack &Stack) {
auto &IDPair = OM[V];
assert(IDPair.first && "Unmapped value");
if (IDPair.second)
// Already predicted.
return;
// Do the actual prediction.
IDPair.second = true;
if (!V->use_empty() && std::next(V->use_begin()) != V->use_end())
predictValueUseListOrderImpl(V, F, IDPair.first, OM, Stack);
// Recursive descent into constants.
if (const Constant *C = dyn_cast<Constant>(V))
if (C->getNumOperands()) // Visit GlobalValues.
for (const Value *Op : C->operands())
if (isa<Constant>(Op)) // Visit GlobalValues.
predictValueUseListOrder(Op, F, OM, Stack);
}
static UseListOrderStack predictUseListOrder(const Module *M) {
OrderMap OM = orderModule(M);
// Use-list orders need to be serialized after all the users have been added
// to a value, or else the shuffles will be incomplete. Store them per
// function in a stack.
//
// Aside from function order, the order of values doesn't matter much here.
UseListOrderStack Stack;
// We want to visit the functions backward now so we can list function-local
// constants in the last Function they're used in. Module-level constants
// have already been visited above.
for (const Function &F : make_range(M->rbegin(), M->rend())) {
if (F.isDeclaration())
continue;
for (const BasicBlock &BB : F)
predictValueUseListOrder(&BB, &F, OM, Stack);
for (const Argument &A : F.args())
predictValueUseListOrder(&A, &F, OM, Stack);
for (const BasicBlock &BB : F)
for (const Instruction &I : BB)
for (const Value *Op : I.operands())
if (isa<Constant>(*Op) || isa<InlineAsm>(*Op)) // Visit GlobalValues.
predictValueUseListOrder(Op, &F, OM, Stack);
for (const BasicBlock &BB : F)
for (const Instruction &I : BB)
predictValueUseListOrder(&I, &F, OM, Stack);
}
// Visit globals last.
for (const GlobalVariable &G : M->globals())
predictValueUseListOrder(&G, nullptr, OM, Stack);
for (const Function &F : *M)
predictValueUseListOrder(&F, nullptr, OM, Stack);
for (const GlobalAlias &A : M->aliases())
predictValueUseListOrder(&A, nullptr, OM, Stack);
for (const GlobalIFunc &I : M->ifuncs())
predictValueUseListOrder(&I, nullptr, OM, Stack);
for (const GlobalVariable &G : M->globals())
if (G.hasInitializer())
predictValueUseListOrder(G.getInitializer(), nullptr, OM, Stack);
for (const GlobalAlias &A : M->aliases())
predictValueUseListOrder(A.getAliasee(), nullptr, OM, Stack);
for (const GlobalIFunc &I : M->ifuncs())
predictValueUseListOrder(I.getResolver(), nullptr, OM, Stack);
for (const Function &F : *M)
for (const Use &U : F.operands())
predictValueUseListOrder(U.get(), nullptr, OM, Stack);
return Stack;
}
static const Module *getModuleFromVal(const Value *V) {
if (const Argument *MA = dyn_cast<Argument>(V))
return MA->getParent() ? MA->getParent()->getParent() : nullptr;
if (const BasicBlock *BB = dyn_cast<BasicBlock>(V))
return BB->getParent() ? BB->getParent()->getParent() : nullptr;
if (const Instruction *I = dyn_cast<Instruction>(V)) {
const Function *M = I->getParent() ? I->getParent()->getParent() : nullptr;
return M ? M->getParent() : nullptr;
}
2011-09-30 21:48:58 +02:00
if (const GlobalValue *GV = dyn_cast<GlobalValue>(V))
return GV->getParent();
IR: Make Metadata::print() reliable and useful Replumb the `AsmWriter` so that `Metadata::print()` is generally useful. (Similarly change `Metadata::printAsOperand()`.) - `SlotTracker` now has a mode where all metadata will be correctly numbered when initializing a `Module`. Normally, `Metadata` only referenced from within `Function`s gets numbered when the `Function` is incorporated. - `Metadata::print()` and `Metadata::printAsOperand()` (and `Metadata::dump()`) now take an optional `Module` argument. When provided, `SlotTracker` is initialized with the new mode, and the numbering will be complete and consistent for all calls to `print()`. - `Value::print()` uses the new `SlotTracker` mode when printing intrinsics with `MDNode` operands, `MetadataAsValue` operands, or the bodies of functions. Thus, metadata numbering will be consistent between calls to `Metadata::print()` and `Value::print()`. - `Metadata::print()` (and `Metadata::dump()`) now print the full definition of `MDNode`s: !5 = !{!6, !"abc", !7} This matches behaviour for `Value::print()`, which includes the name of instructions. - Updated call sites in `Verifier` to call `print()` instead of `printAsOperand()`. All this, so that `Verifier` can print out useful failure messages that involve `Metadata` for PR22777. Note that `Metadata::printAsOperand()` previously took an optional `bool` and `Module` operand. The former was cargo-culted from `Value::printAsOperand()` and wasn't doing anything useful. The latter didn't give consistent results (without the new `SlotTracker` mode). llvm-svn: 232275
2015-03-14 21:19:36 +01:00
if (const auto *MAV = dyn_cast<MetadataAsValue>(V)) {
for (const User *U : MAV->users())
if (isa<Instruction>(U))
if (const Module *M = getModuleFromVal(U))
return M;
return nullptr;
}
return nullptr;
}
static void PrintCallingConv(unsigned cc, raw_ostream &Out) {
switch (cc) {
default: Out << "cc" << cc; break;
case CallingConv::Fast: Out << "fastcc"; break;
case CallingConv::Cold: Out << "coldcc"; break;
case CallingConv::WebKit_JS: Out << "webkit_jscc"; break;
case CallingConv::AnyReg: Out << "anyregcc"; break;
case CallingConv::PreserveMost: Out << "preserve_mostcc"; break;
case CallingConv::PreserveAll: Out << "preserve_allcc"; break;
case CallingConv::CXX_FAST_TLS: Out << "cxx_fast_tlscc"; break;
case CallingConv::GHC: Out << "ghccc"; break;
case CallingConv::X86_StdCall: Out << "x86_stdcallcc"; break;
case CallingConv::X86_FastCall: Out << "x86_fastcallcc"; break;
case CallingConv::X86_ThisCall: Out << "x86_thiscallcc"; break;
case CallingConv::X86_RegCall: Out << "x86_regcallcc"; break;
case CallingConv::X86_VectorCall:Out << "x86_vectorcallcc"; break;
case CallingConv::Intel_OCL_BI: Out << "intel_ocl_bicc"; break;
case CallingConv::ARM_APCS: Out << "arm_apcscc"; break;
case CallingConv::ARM_AAPCS: Out << "arm_aapcscc"; break;
case CallingConv::ARM_AAPCS_VFP: Out << "arm_aapcs_vfpcc"; break;
case CallingConv::MSP430_INTR: Out << "msp430_intrcc"; break;
case CallingConv::AVR_INTR: Out << "avr_intrcc "; break;
case CallingConv::AVR_SIGNAL: Out << "avr_signalcc "; break;
case CallingConv::PTX_Kernel: Out << "ptx_kernel"; break;
case CallingConv::PTX_Device: Out << "ptx_device"; break;
case CallingConv::X86_64_SysV: Out << "x86_64_sysvcc"; break;
case CallingConv::Win64: Out << "win64cc"; break;
case CallingConv::SPIR_FUNC: Out << "spir_func"; break;
case CallingConv::SPIR_KERNEL: Out << "spir_kernel"; break;
case CallingConv::Swift: Out << "swiftcc"; break;
case CallingConv::X86_INTR: Out << "x86_intrcc"; break;
case CallingConv::HHVM: Out << "hhvmcc"; break;
case CallingConv::HHVM_C: Out << "hhvm_ccc"; break;
case CallingConv::AMDGPU_VS: Out << "amdgpu_vs"; break;
case CallingConv::AMDGPU_LS: Out << "amdgpu_ls"; break;
case CallingConv::AMDGPU_HS: Out << "amdgpu_hs"; break;
case CallingConv::AMDGPU_ES: Out << "amdgpu_es"; break;
case CallingConv::AMDGPU_GS: Out << "amdgpu_gs"; break;
case CallingConv::AMDGPU_PS: Out << "amdgpu_ps"; break;
case CallingConv::AMDGPU_CS: Out << "amdgpu_cs"; break;
case CallingConv::AMDGPU_KERNEL: Out << "amdgpu_kernel"; break;
}
}
enum PrefixType {
GlobalPrefix,
ComdatPrefix,
LabelPrefix,
LocalPrefix,
NoPrefix
};
void llvm::printLLVMNameWithoutPrefix(raw_ostream &OS, StringRef Name) {
assert(!Name.empty() && "Cannot get empty name!");
// Scan the name to see if it needs quotes first.
bool NeedsQuotes = isdigit(static_cast<unsigned char>(Name[0]));
if (!NeedsQuotes) {
for (unsigned i = 0, e = Name.size(); i != e; ++i) {
// By making this unsigned, the value passed in to isalnum will always be
// in the range 0-255. This is important when building with MSVC because
// its implementation will assert. This situation can arise when dealing
// with UTF-8 multibyte characters.
unsigned char C = Name[i];
if (!isalnum(static_cast<unsigned char>(C)) && C != '-' && C != '.' &&
C != '_') {
NeedsQuotes = true;
break;
}
}
}
// If we didn't need any quotes, just write out the name in one blast.
if (!NeedsQuotes) {
OS << Name;
return;
}
// Okay, we need quotes. Output the quotes and escape any scary characters as
// needed.
OS << '"';
PrintEscapedString(Name, OS);
OS << '"';
}
/// Turn the specified name into an 'LLVM name', which is either prefixed with %
/// (if the string only contains simple characters) or is surrounded with ""'s
/// (if it has special chars in it). Print it out.
static void PrintLLVMName(raw_ostream &OS, StringRef Name, PrefixType Prefix) {
switch (Prefix) {
case NoPrefix:
break;
case GlobalPrefix:
OS << '@';
break;
case ComdatPrefix:
OS << '$';
break;
case LabelPrefix:
break;
case LocalPrefix:
OS << '%';
break;
}
printLLVMNameWithoutPrefix(OS, Name);
}
/// Turn the specified name into an 'LLVM name', which is either prefixed with %
/// (if the string only contains simple characters) or is surrounded with ""'s
/// (if it has special chars in it). Print it out.
static void PrintLLVMName(raw_ostream &OS, const Value *V) {
PrintLLVMName(OS, V->getName(),
isa<GlobalValue>(V) ? GlobalPrefix : LocalPrefix);
}
namespace {
class TypePrinting {
public:
[MIR] Making MIR Printing, opt -dot-cfg, and -debug printing faster Value::printAsOperand has been scanning the entire module just to print a single value as an operand, regardless being asked to print a type or not at all, and regardless really needing to scan the module to print a type. It made some of the users of the method exceptionally slow on large IR-modules (or large MIR-files with large IR-modules embedded). This patch defers scanning a module looking for struct types, mostly numbered struct types, as much as possible, speeding up those users w/o changing any APIs at all. See speedup examples below: Release Build: # 83 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 133 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output Release + Asserts Build: # 95 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 146 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output # 1096 seconds -> 553 seconds time ./bin/llc -debug-only=isel -fast-isel=false -stop-after=isel \ sqlite3.O0.ll -o /dev/null 2> err where sqlite3.O0.ll is non-optimized IR produced from sqlite-amalgamation (http://sqlite.org/download.html), which is entire SQLite3 implementation in a single C-file. Benchmarked on 4-cores / 8 threads PCI-E SSD iMac running macOS Reviewers: dexonsmith, bkramer, void, chandlerc, aditya_nandakumar, dsanders, qcolombet, Reviewed By: bogner Subscribers: thegameg, llvm-commits Differential Revision: https://reviews.llvm.org/D44132 llvm-svn: 328246
2018-03-22 22:29:07 +01:00
TypePrinting(const Module *M = nullptr) : DeferredM(M) {}
TypePrinting(const TypePrinting &) = delete;
TypePrinting &operator=(const TypePrinting &) = delete;
[MIR] Making MIR Printing, opt -dot-cfg, and -debug printing faster Value::printAsOperand has been scanning the entire module just to print a single value as an operand, regardless being asked to print a type or not at all, and regardless really needing to scan the module to print a type. It made some of the users of the method exceptionally slow on large IR-modules (or large MIR-files with large IR-modules embedded). This patch defers scanning a module looking for struct types, mostly numbered struct types, as much as possible, speeding up those users w/o changing any APIs at all. See speedup examples below: Release Build: # 83 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 133 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output Release + Asserts Build: # 95 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 146 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output # 1096 seconds -> 553 seconds time ./bin/llc -debug-only=isel -fast-isel=false -stop-after=isel \ sqlite3.O0.ll -o /dev/null 2> err where sqlite3.O0.ll is non-optimized IR produced from sqlite-amalgamation (http://sqlite.org/download.html), which is entire SQLite3 implementation in a single C-file. Benchmarked on 4-cores / 8 threads PCI-E SSD iMac running macOS Reviewers: dexonsmith, bkramer, void, chandlerc, aditya_nandakumar, dsanders, qcolombet, Reviewed By: bogner Subscribers: thegameg, llvm-commits Differential Revision: https://reviews.llvm.org/D44132 llvm-svn: 328246
2018-03-22 22:29:07 +01:00
/// The named types that are used by the current module.
TypeFinder &getNamedTypes();
/// The numbered types, number to type mapping.
std::vector<StructType *> &getNumberedTypes();
bool empty();
void print(Type *Ty, raw_ostream &OS);
void printStructBody(StructType *Ty, raw_ostream &OS);
[MIR] Making MIR Printing, opt -dot-cfg, and -debug printing faster Value::printAsOperand has been scanning the entire module just to print a single value as an operand, regardless being asked to print a type or not at all, and regardless really needing to scan the module to print a type. It made some of the users of the method exceptionally slow on large IR-modules (or large MIR-files with large IR-modules embedded). This patch defers scanning a module looking for struct types, mostly numbered struct types, as much as possible, speeding up those users w/o changing any APIs at all. See speedup examples below: Release Build: # 83 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 133 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output Release + Asserts Build: # 95 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 146 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output # 1096 seconds -> 553 seconds time ./bin/llc -debug-only=isel -fast-isel=false -stop-after=isel \ sqlite3.O0.ll -o /dev/null 2> err where sqlite3.O0.ll is non-optimized IR produced from sqlite-amalgamation (http://sqlite.org/download.html), which is entire SQLite3 implementation in a single C-file. Benchmarked on 4-cores / 8 threads PCI-E SSD iMac running macOS Reviewers: dexonsmith, bkramer, void, chandlerc, aditya_nandakumar, dsanders, qcolombet, Reviewed By: bogner Subscribers: thegameg, llvm-commits Differential Revision: https://reviews.llvm.org/D44132 llvm-svn: 328246
2018-03-22 22:29:07 +01:00
private:
void incorporateTypes();
/// A module to process lazily when needed. Set to nullptr as soon as used.
const Module *DeferredM;
TypeFinder NamedTypes;
// The numbered types, along with their value.
DenseMap<StructType *, unsigned> Type2Number;
std::vector<StructType *> NumberedTypes;
};
} // end anonymous namespace
[MIR] Making MIR Printing, opt -dot-cfg, and -debug printing faster Value::printAsOperand has been scanning the entire module just to print a single value as an operand, regardless being asked to print a type or not at all, and regardless really needing to scan the module to print a type. It made some of the users of the method exceptionally slow on large IR-modules (or large MIR-files with large IR-modules embedded). This patch defers scanning a module looking for struct types, mostly numbered struct types, as much as possible, speeding up those users w/o changing any APIs at all. See speedup examples below: Release Build: # 83 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 133 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output Release + Asserts Build: # 95 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 146 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output # 1096 seconds -> 553 seconds time ./bin/llc -debug-only=isel -fast-isel=false -stop-after=isel \ sqlite3.O0.ll -o /dev/null 2> err where sqlite3.O0.ll is non-optimized IR produced from sqlite-amalgamation (http://sqlite.org/download.html), which is entire SQLite3 implementation in a single C-file. Benchmarked on 4-cores / 8 threads PCI-E SSD iMac running macOS Reviewers: dexonsmith, bkramer, void, chandlerc, aditya_nandakumar, dsanders, qcolombet, Reviewed By: bogner Subscribers: thegameg, llvm-commits Differential Revision: https://reviews.llvm.org/D44132 llvm-svn: 328246
2018-03-22 22:29:07 +01:00
TypeFinder &TypePrinting::getNamedTypes() {
incorporateTypes();
return NamedTypes;
}
std::vector<StructType *> &TypePrinting::getNumberedTypes() {
incorporateTypes();
// We know all the numbers that each type is used and we know that it is a
// dense assignment. Convert the map to an index table, if it's not done
// already (judging from the sizes):
if (NumberedTypes.size() == Type2Number.size())
return NumberedTypes;
NumberedTypes.resize(Type2Number.size());
for (const auto &P : Type2Number) {
assert(P.second < NumberedTypes.size() && "Didn't get a dense numbering?");
assert(!NumberedTypes[P.second] && "Didn't get a unique numbering?");
NumberedTypes[P.second] = P.first;
}
return NumberedTypes;
}
bool TypePrinting::empty() {
incorporateTypes();
return NamedTypes.empty() && Type2Number.empty();
}
void TypePrinting::incorporateTypes() {
if (!DeferredM)
return;
NamedTypes.run(*DeferredM, false);
DeferredM = nullptr;
2011-09-30 21:48:58 +02:00
// The list of struct types we got back includes all the struct types, split
// the unnamed ones out to a numbering and remove the anonymous structs.
unsigned NextNumber = 0;
2011-09-30 21:48:58 +02:00
std::vector<StructType*>::iterator NextToUse = NamedTypes.begin(), I, E;
for (I = NamedTypes.begin(), E = NamedTypes.end(); I != E; ++I) {
StructType *STy = *I;
2011-09-30 21:48:58 +02:00
// Ignore anonymous types.
if (STy->isLiteral())
continue;
2011-09-30 21:48:58 +02:00
if (STy->getName().empty())
[MIR] Making MIR Printing, opt -dot-cfg, and -debug printing faster Value::printAsOperand has been scanning the entire module just to print a single value as an operand, regardless being asked to print a type or not at all, and regardless really needing to scan the module to print a type. It made some of the users of the method exceptionally slow on large IR-modules (or large MIR-files with large IR-modules embedded). This patch defers scanning a module looking for struct types, mostly numbered struct types, as much as possible, speeding up those users w/o changing any APIs at all. See speedup examples below: Release Build: # 83 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 133 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output Release + Asserts Build: # 95 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 146 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output # 1096 seconds -> 553 seconds time ./bin/llc -debug-only=isel -fast-isel=false -stop-after=isel \ sqlite3.O0.ll -o /dev/null 2> err where sqlite3.O0.ll is non-optimized IR produced from sqlite-amalgamation (http://sqlite.org/download.html), which is entire SQLite3 implementation in a single C-file. Benchmarked on 4-cores / 8 threads PCI-E SSD iMac running macOS Reviewers: dexonsmith, bkramer, void, chandlerc, aditya_nandakumar, dsanders, qcolombet, Reviewed By: bogner Subscribers: thegameg, llvm-commits Differential Revision: https://reviews.llvm.org/D44132 llvm-svn: 328246
2018-03-22 22:29:07 +01:00
Type2Number[STy] = NextNumber++;
else
*NextToUse++ = STy;
}
2011-09-30 21:48:58 +02:00
NamedTypes.erase(NextToUse, NamedTypes.end());
}
[MIR] Making MIR Printing, opt -dot-cfg, and -debug printing faster Value::printAsOperand has been scanning the entire module just to print a single value as an operand, regardless being asked to print a type or not at all, and regardless really needing to scan the module to print a type. It made some of the users of the method exceptionally slow on large IR-modules (or large MIR-files with large IR-modules embedded). This patch defers scanning a module looking for struct types, mostly numbered struct types, as much as possible, speeding up those users w/o changing any APIs at all. See speedup examples below: Release Build: # 83 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 133 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output Release + Asserts Build: # 95 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 146 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output # 1096 seconds -> 553 seconds time ./bin/llc -debug-only=isel -fast-isel=false -stop-after=isel \ sqlite3.O0.ll -o /dev/null 2> err where sqlite3.O0.ll is non-optimized IR produced from sqlite-amalgamation (http://sqlite.org/download.html), which is entire SQLite3 implementation in a single C-file. Benchmarked on 4-cores / 8 threads PCI-E SSD iMac running macOS Reviewers: dexonsmith, bkramer, void, chandlerc, aditya_nandakumar, dsanders, qcolombet, Reviewed By: bogner Subscribers: thegameg, llvm-commits Differential Revision: https://reviews.llvm.org/D44132 llvm-svn: 328246
2018-03-22 22:29:07 +01:00
/// Write the specified type to the specified raw_ostream, making use of type
/// names or up references to shorten the type name where possible.
void TypePrinting::print(Type *Ty, raw_ostream &OS) {
switch (Ty->getTypeID()) {
2013-12-07 03:27:52 +01:00
case Type::VoidTyID: OS << "void"; return;
case Type::HalfTyID: OS << "half"; return;
case Type::FloatTyID: OS << "float"; return;
case Type::DoubleTyID: OS << "double"; return;
case Type::X86_FP80TyID: OS << "x86_fp80"; return;
case Type::FP128TyID: OS << "fp128"; return;
case Type::PPC_FP128TyID: OS << "ppc_fp128"; return;
case Type::LabelTyID: OS << "label"; return;
case Type::MetadataTyID: OS << "metadata"; return;
case Type::X86_MMXTyID: OS << "x86_mmx"; return;
[IR] Add token types This introduces the basic functionality to support "token types". The motivation stems from the need to perform operations on a Value whose provenance cannot be obscured. There are several applications for such a type but my immediate motivation stems from WinEH. Our personality routine enforces a single-entry - single-exit regime for cleanups. After several rounds of optimizations, we may be left with a terminator whose "cleanup-entry block" is not entirely clear because control flow has merged two cleanups together. We have experimented with using labels as operands inside of instructions which are not terminators to indicate where we came from but found that LLVM does not expect such exotic uses of BasicBlocks. Instead, we can use this new type to clearly associate the "entry point" and "exit point" of our cleanup. This is done by having the cleanuppad yield a Token and consuming it at the cleanupret. The token type makes it impossible to obscure or otherwise hide the Value, making it trivial to track the relationship between the two points. What is the burden to the optimizer? Well, it turns out we have already paid down this cost by accepting that there are certain calls that we are not permitted to duplicate, optimizations have to watch out for such instructions anyway. There are additional places in the optimizer that we will probably have to update but early examination has given me the impression that this will not be heroic. Differential Revision: http://reviews.llvm.org/D11861 llvm-svn: 245029
2015-08-14 07:09:07 +02:00
case Type::TokenTyID: OS << "token"; return;
case Type::IntegerTyID:
OS << 'i' << cast<IntegerType>(Ty)->getBitWidth();
return;
case Type::FunctionTyID: {
FunctionType *FTy = cast<FunctionType>(Ty);
print(FTy->getReturnType(), OS);
OS << " (";
for (FunctionType::param_iterator I = FTy->param_begin(),
E = FTy->param_end(); I != E; ++I) {
if (I != FTy->param_begin())
OS << ", ";
print(*I, OS);
}
if (FTy->isVarArg()) {
if (FTy->getNumParams()) OS << ", ";
OS << "...";
}
OS << ')';
return;
}
case Type::StructTyID: {
StructType *STy = cast<StructType>(Ty);
2011-09-30 21:48:58 +02:00
if (STy->isLiteral())
return printStructBody(STy, OS);
if (!STy->getName().empty())
return PrintLLVMName(OS, STy->getName(), LocalPrefix);
2011-09-30 21:48:58 +02:00
[MIR] Making MIR Printing, opt -dot-cfg, and -debug printing faster Value::printAsOperand has been scanning the entire module just to print a single value as an operand, regardless being asked to print a type or not at all, and regardless really needing to scan the module to print a type. It made some of the users of the method exceptionally slow on large IR-modules (or large MIR-files with large IR-modules embedded). This patch defers scanning a module looking for struct types, mostly numbered struct types, as much as possible, speeding up those users w/o changing any APIs at all. See speedup examples below: Release Build: # 83 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 133 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output Release + Asserts Build: # 95 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 146 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output # 1096 seconds -> 553 seconds time ./bin/llc -debug-only=isel -fast-isel=false -stop-after=isel \ sqlite3.O0.ll -o /dev/null 2> err where sqlite3.O0.ll is non-optimized IR produced from sqlite-amalgamation (http://sqlite.org/download.html), which is entire SQLite3 implementation in a single C-file. Benchmarked on 4-cores / 8 threads PCI-E SSD iMac running macOS Reviewers: dexonsmith, bkramer, void, chandlerc, aditya_nandakumar, dsanders, qcolombet, Reviewed By: bogner Subscribers: thegameg, llvm-commits Differential Revision: https://reviews.llvm.org/D44132 llvm-svn: 328246
2018-03-22 22:29:07 +01:00
incorporateTypes();
const auto I = Type2Number.find(STy);
if (I != Type2Number.end())
OS << '%' << I->second;
else // Not enumerated, print the hex address.
OS << "%\"type " << STy << '\"';
return;
}
case Type::PointerTyID: {
PointerType *PTy = cast<PointerType>(Ty);
print(PTy->getElementType(), OS);
if (unsigned AddressSpace = PTy->getAddressSpace())
OS << " addrspace(" << AddressSpace << ')';
OS << '*';
return;
}
case Type::ArrayTyID: {
ArrayType *ATy = cast<ArrayType>(Ty);
OS << '[' << ATy->getNumElements() << " x ";
print(ATy->getElementType(), OS);
OS << ']';
return;
}
case Type::VectorTyID: {
VectorType *PTy = cast<VectorType>(Ty);
OS << "<" << PTy->getNumElements() << " x ";
print(PTy->getElementType(), OS);
OS << '>';
return;
}
}
2013-12-07 03:27:52 +01:00
llvm_unreachable("Invalid TypeID");
}
void TypePrinting::printStructBody(StructType *STy, raw_ostream &OS) {
if (STy->isOpaque()) {
OS << "opaque";
return;
}
2011-09-30 21:48:58 +02:00
if (STy->isPacked())
OS << '<';
2011-09-30 21:48:58 +02:00
if (STy->getNumElements() == 0) {
OS << "{}";
} else {
StructType::element_iterator I = STy->element_begin();
OS << "{ ";
print(*I++, OS);
for (StructType::element_iterator E = STy->element_end(); I != E; ++I) {
OS << ", ";
print(*I, OS);
}
2011-09-30 21:48:58 +02:00
OS << " }";
}
if (STy->isPacked())
OS << '>';
}
namespace llvm {
//===----------------------------------------------------------------------===//
// SlotTracker Class: Enumerate slot numbers for unnamed values
//===----------------------------------------------------------------------===//
/// This class provides computation of slot numbers for LLVM Assembly writing.
///
class SlotTracker {
public:
/// ValueMap - A mapping of Values to slot numbers.
using ValueMap = DenseMap<const Value *, unsigned>;
private:
/// TheModule - The module for which we are holding slot numbers.
const Module* TheModule;
/// TheFunction - The function for which we are holding slot numbers.
const Function* TheFunction = nullptr;
bool FunctionProcessed = false;
IR: Make Metadata::print() reliable and useful Replumb the `AsmWriter` so that `Metadata::print()` is generally useful. (Similarly change `Metadata::printAsOperand()`.) - `SlotTracker` now has a mode where all metadata will be correctly numbered when initializing a `Module`. Normally, `Metadata` only referenced from within `Function`s gets numbered when the `Function` is incorporated. - `Metadata::print()` and `Metadata::printAsOperand()` (and `Metadata::dump()`) now take an optional `Module` argument. When provided, `SlotTracker` is initialized with the new mode, and the numbering will be complete and consistent for all calls to `print()`. - `Value::print()` uses the new `SlotTracker` mode when printing intrinsics with `MDNode` operands, `MetadataAsValue` operands, or the bodies of functions. Thus, metadata numbering will be consistent between calls to `Metadata::print()` and `Value::print()`. - `Metadata::print()` (and `Metadata::dump()`) now print the full definition of `MDNode`s: !5 = !{!6, !"abc", !7} This matches behaviour for `Value::print()`, which includes the name of instructions. - Updated call sites in `Verifier` to call `print()` instead of `printAsOperand()`. All this, so that `Verifier` can print out useful failure messages that involve `Metadata` for PR22777. Note that `Metadata::printAsOperand()` previously took an optional `bool` and `Module` operand. The former was cargo-culted from `Value::printAsOperand()` and wasn't doing anything useful. The latter didn't give consistent results (without the new `SlotTracker` mode). llvm-svn: 232275
2015-03-14 21:19:36 +01:00
bool ShouldInitializeAllMetadata;
/// mMap - The slot map for the module level data.
ValueMap mMap;
unsigned mNext = 0;
/// fMap - The slot map for the function level data.
ValueMap fMap;
unsigned fNext = 0;
/// mdnMap - Map for MDNodes.
DenseMap<const MDNode*, unsigned> mdnMap;
unsigned mdnNext = 0;
/// asMap - The slot map for attribute sets.
DenseMap<AttributeSet, unsigned> asMap;
unsigned asNext = 0;
public:
IR: Make Metadata::print() reliable and useful Replumb the `AsmWriter` so that `Metadata::print()` is generally useful. (Similarly change `Metadata::printAsOperand()`.) - `SlotTracker` now has a mode where all metadata will be correctly numbered when initializing a `Module`. Normally, `Metadata` only referenced from within `Function`s gets numbered when the `Function` is incorporated. - `Metadata::print()` and `Metadata::printAsOperand()` (and `Metadata::dump()`) now take an optional `Module` argument. When provided, `SlotTracker` is initialized with the new mode, and the numbering will be complete and consistent for all calls to `print()`. - `Value::print()` uses the new `SlotTracker` mode when printing intrinsics with `MDNode` operands, `MetadataAsValue` operands, or the bodies of functions. Thus, metadata numbering will be consistent between calls to `Metadata::print()` and `Value::print()`. - `Metadata::print()` (and `Metadata::dump()`) now print the full definition of `MDNode`s: !5 = !{!6, !"abc", !7} This matches behaviour for `Value::print()`, which includes the name of instructions. - Updated call sites in `Verifier` to call `print()` instead of `printAsOperand()`. All this, so that `Verifier` can print out useful failure messages that involve `Metadata` for PR22777. Note that `Metadata::printAsOperand()` previously took an optional `bool` and `Module` operand. The former was cargo-culted from `Value::printAsOperand()` and wasn't doing anything useful. The latter didn't give consistent results (without the new `SlotTracker` mode). llvm-svn: 232275
2015-03-14 21:19:36 +01:00
/// Construct from a module.
///
/// If \c ShouldInitializeAllMetadata, initializes all metadata in all
/// functions, giving correct numbering for metadata referenced only from
/// within a function (even if no functions have been initialized).
explicit SlotTracker(const Module *M,
bool ShouldInitializeAllMetadata = false);
/// Construct from a function, starting out in incorp state.
IR: Make Metadata::print() reliable and useful Replumb the `AsmWriter` so that `Metadata::print()` is generally useful. (Similarly change `Metadata::printAsOperand()`.) - `SlotTracker` now has a mode where all metadata will be correctly numbered when initializing a `Module`. Normally, `Metadata` only referenced from within `Function`s gets numbered when the `Function` is incorporated. - `Metadata::print()` and `Metadata::printAsOperand()` (and `Metadata::dump()`) now take an optional `Module` argument. When provided, `SlotTracker` is initialized with the new mode, and the numbering will be complete and consistent for all calls to `print()`. - `Value::print()` uses the new `SlotTracker` mode when printing intrinsics with `MDNode` operands, `MetadataAsValue` operands, or the bodies of functions. Thus, metadata numbering will be consistent between calls to `Metadata::print()` and `Value::print()`. - `Metadata::print()` (and `Metadata::dump()`) now print the full definition of `MDNode`s: !5 = !{!6, !"abc", !7} This matches behaviour for `Value::print()`, which includes the name of instructions. - Updated call sites in `Verifier` to call `print()` instead of `printAsOperand()`. All this, so that `Verifier` can print out useful failure messages that involve `Metadata` for PR22777. Note that `Metadata::printAsOperand()` previously took an optional `bool` and `Module` operand. The former was cargo-culted from `Value::printAsOperand()` and wasn't doing anything useful. The latter didn't give consistent results (without the new `SlotTracker` mode). llvm-svn: 232275
2015-03-14 21:19:36 +01:00
///
/// If \c ShouldInitializeAllMetadata, initializes all metadata in all
/// functions, giving correct numbering for metadata referenced only from
/// within a function (even if no functions have been initialized).
explicit SlotTracker(const Function *F,
bool ShouldInitializeAllMetadata = false);
SlotTracker(const SlotTracker &) = delete;
SlotTracker &operator=(const SlotTracker &) = delete;
/// Return the slot number of the specified value in it's type
/// plane. If something is not in the SlotTracker, return -1.
int getLocalSlot(const Value *V);
int getGlobalSlot(const GlobalValue *V);
int getMetadataSlot(const MDNode *N);
int getAttributeGroupSlot(AttributeSet AS);
/// If you'd like to deal with a function instead of just a module, use
/// this method to get its data into the SlotTracker.
void incorporateFunction(const Function *F) {
TheFunction = F;
FunctionProcessed = false;
}
const Function *getFunction() const { return TheFunction; }
/// After calling incorporateFunction, use this method to remove the
/// most recently incorporated function from the SlotTracker. This
/// will reset the state of the machine back to just the module contents.
void purgeFunction();
/// MDNode map iterators.
using mdn_iterator = DenseMap<const MDNode*, unsigned>::iterator;
mdn_iterator mdn_begin() { return mdnMap.begin(); }
mdn_iterator mdn_end() { return mdnMap.end(); }
unsigned mdn_size() const { return mdnMap.size(); }
bool mdn_empty() const { return mdnMap.empty(); }
/// AttributeSet map iterators.
using as_iterator = DenseMap<AttributeSet, unsigned>::iterator;
as_iterator as_begin() { return asMap.begin(); }
as_iterator as_end() { return asMap.end(); }
unsigned as_size() const { return asMap.size(); }
bool as_empty() const { return asMap.empty(); }
/// This function does the actual initialization.
inline void initialize();
// Implementation Details
private:
/// CreateModuleSlot - Insert the specified GlobalValue* into the slot table.
void CreateModuleSlot(const GlobalValue *V);
/// CreateMetadataSlot - Insert the specified MDNode* into the slot table.
void CreateMetadataSlot(const MDNode *N);
/// CreateFunctionSlot - Insert the specified Value* into the slot table.
void CreateFunctionSlot(const Value *V);
/// Insert the specified AttributeSet into the slot table.
void CreateAttributeSetSlot(AttributeSet AS);
/// Add all of the module level global variables (and their initializers)
/// and function declarations, but not the contents of those functions.
void processModule();
/// Add all of the functions arguments, basic blocks, and instructions.
void processFunction();
/// Add the metadata directly attached to a GlobalObject.
void processGlobalObjectMetadata(const GlobalObject &GO);
IR: Make Metadata::print() reliable and useful Replumb the `AsmWriter` so that `Metadata::print()` is generally useful. (Similarly change `Metadata::printAsOperand()`.) - `SlotTracker` now has a mode where all metadata will be correctly numbered when initializing a `Module`. Normally, `Metadata` only referenced from within `Function`s gets numbered when the `Function` is incorporated. - `Metadata::print()` and `Metadata::printAsOperand()` (and `Metadata::dump()`) now take an optional `Module` argument. When provided, `SlotTracker` is initialized with the new mode, and the numbering will be complete and consistent for all calls to `print()`. - `Value::print()` uses the new `SlotTracker` mode when printing intrinsics with `MDNode` operands, `MetadataAsValue` operands, or the bodies of functions. Thus, metadata numbering will be consistent between calls to `Metadata::print()` and `Value::print()`. - `Metadata::print()` (and `Metadata::dump()`) now print the full definition of `MDNode`s: !5 = !{!6, !"abc", !7} This matches behaviour for `Value::print()`, which includes the name of instructions. - Updated call sites in `Verifier` to call `print()` instead of `printAsOperand()`. All this, so that `Verifier` can print out useful failure messages that involve `Metadata` for PR22777. Note that `Metadata::printAsOperand()` previously took an optional `bool` and `Module` operand. The former was cargo-culted from `Value::printAsOperand()` and wasn't doing anything useful. The latter didn't give consistent results (without the new `SlotTracker` mode). llvm-svn: 232275
2015-03-14 21:19:36 +01:00
/// Add all of the metadata from a function.
void processFunctionMetadata(const Function &F);
/// Add all of the metadata from an instruction.
void processInstructionMetadata(const Instruction &I);
};
} // end namespace llvm
ModuleSlotTracker::ModuleSlotTracker(SlotTracker &Machine, const Module *M,
const Function *F)
: M(M), F(F), Machine(&Machine) {}
ModuleSlotTracker::ModuleSlotTracker(const Module *M,
bool ShouldInitializeAllMetadata)
: ShouldCreateStorage(M),
ShouldInitializeAllMetadata(ShouldInitializeAllMetadata), M(M) {}
ModuleSlotTracker::~ModuleSlotTracker() = default;
SlotTracker *ModuleSlotTracker::getMachine() {
if (!ShouldCreateStorage)
return Machine;
ShouldCreateStorage = false;
MachineStorage =
llvm::make_unique<SlotTracker>(M, ShouldInitializeAllMetadata);
Machine = MachineStorage.get();
return Machine;
}
void ModuleSlotTracker::incorporateFunction(const Function &F) {
// Using getMachine() may lazily create the slot tracker.
if (!getMachine())
return;
// Nothing to do if this is the right function already.
if (this->F == &F)
return;
if (this->F)
Machine->purgeFunction();
Machine->incorporateFunction(&F);
this->F = &F;
}
int ModuleSlotTracker::getLocalSlot(const Value *V) {
assert(F && "No function incorporated");
return Machine->getLocalSlot(V);
}
static SlotTracker *createSlotTracker(const Value *V) {
if (const Argument *FA = dyn_cast<Argument>(V))
return new SlotTracker(FA->getParent());
if (const Instruction *I = dyn_cast<Instruction>(V))
if (I->getParent())
return new SlotTracker(I->getParent()->getParent());
if (const BasicBlock *BB = dyn_cast<BasicBlock>(V))
return new SlotTracker(BB->getParent());
if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(V))
return new SlotTracker(GV->getParent());
if (const GlobalAlias *GA = dyn_cast<GlobalAlias>(V))
return new SlotTracker(GA->getParent());
if (const GlobalIFunc *GIF = dyn_cast<GlobalIFunc>(V))
return new SlotTracker(GIF->getParent());
if (const Function *Func = dyn_cast<Function>(V))
return new SlotTracker(Func);
return nullptr;
}
#if 0
#define ST_DEBUG(X) dbgs() << X
#else
#define ST_DEBUG(X)
#endif
// Module level constructor. Causes the contents of the Module (sans functions)
// to be added to the slot table.
IR: Make Metadata::print() reliable and useful Replumb the `AsmWriter` so that `Metadata::print()` is generally useful. (Similarly change `Metadata::printAsOperand()`.) - `SlotTracker` now has a mode where all metadata will be correctly numbered when initializing a `Module`. Normally, `Metadata` only referenced from within `Function`s gets numbered when the `Function` is incorporated. - `Metadata::print()` and `Metadata::printAsOperand()` (and `Metadata::dump()`) now take an optional `Module` argument. When provided, `SlotTracker` is initialized with the new mode, and the numbering will be complete and consistent for all calls to `print()`. - `Value::print()` uses the new `SlotTracker` mode when printing intrinsics with `MDNode` operands, `MetadataAsValue` operands, or the bodies of functions. Thus, metadata numbering will be consistent between calls to `Metadata::print()` and `Value::print()`. - `Metadata::print()` (and `Metadata::dump()`) now print the full definition of `MDNode`s: !5 = !{!6, !"abc", !7} This matches behaviour for `Value::print()`, which includes the name of instructions. - Updated call sites in `Verifier` to call `print()` instead of `printAsOperand()`. All this, so that `Verifier` can print out useful failure messages that involve `Metadata` for PR22777. Note that `Metadata::printAsOperand()` previously took an optional `bool` and `Module` operand. The former was cargo-culted from `Value::printAsOperand()` and wasn't doing anything useful. The latter didn't give consistent results (without the new `SlotTracker` mode). llvm-svn: 232275
2015-03-14 21:19:36 +01:00
SlotTracker::SlotTracker(const Module *M, bool ShouldInitializeAllMetadata)
: TheModule(M), ShouldInitializeAllMetadata(ShouldInitializeAllMetadata) {}
// Function level constructor. Causes the contents of the Module and the one
// function provided to be added to the slot table.
IR: Make Metadata::print() reliable and useful Replumb the `AsmWriter` so that `Metadata::print()` is generally useful. (Similarly change `Metadata::printAsOperand()`.) - `SlotTracker` now has a mode where all metadata will be correctly numbered when initializing a `Module`. Normally, `Metadata` only referenced from within `Function`s gets numbered when the `Function` is incorporated. - `Metadata::print()` and `Metadata::printAsOperand()` (and `Metadata::dump()`) now take an optional `Module` argument. When provided, `SlotTracker` is initialized with the new mode, and the numbering will be complete and consistent for all calls to `print()`. - `Value::print()` uses the new `SlotTracker` mode when printing intrinsics with `MDNode` operands, `MetadataAsValue` operands, or the bodies of functions. Thus, metadata numbering will be consistent between calls to `Metadata::print()` and `Value::print()`. - `Metadata::print()` (and `Metadata::dump()`) now print the full definition of `MDNode`s: !5 = !{!6, !"abc", !7} This matches behaviour for `Value::print()`, which includes the name of instructions. - Updated call sites in `Verifier` to call `print()` instead of `printAsOperand()`. All this, so that `Verifier` can print out useful failure messages that involve `Metadata` for PR22777. Note that `Metadata::printAsOperand()` previously took an optional `bool` and `Module` operand. The former was cargo-culted from `Value::printAsOperand()` and wasn't doing anything useful. The latter didn't give consistent results (without the new `SlotTracker` mode). llvm-svn: 232275
2015-03-14 21:19:36 +01:00
SlotTracker::SlotTracker(const Function *F, bool ShouldInitializeAllMetadata)
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
: TheModule(F ? F->getParent() : nullptr), TheFunction(F),
ShouldInitializeAllMetadata(ShouldInitializeAllMetadata) {}
inline void SlotTracker::initialize() {
if (TheModule) {
processModule();
TheModule = nullptr; ///< Prevent re-processing next time we're called.
}
if (TheFunction && !FunctionProcessed)
processFunction();
}
// Iterate through all the global variables, functions, and global
// variable initializers and create slots for them.
void SlotTracker::processModule() {
ST_DEBUG("begin processModule!\n");
// Add all of the unnamed global variables to the value table.
for (const GlobalVariable &Var : TheModule->globals()) {
if (!Var.hasName())
CreateModuleSlot(&Var);
processGlobalObjectMetadata(Var);
auto Attrs = Var.getAttributes();
if (Attrs.hasAttributes())
CreateAttributeSetSlot(Attrs);
}
for (const GlobalAlias &A : TheModule->aliases()) {
if (!A.hasName())
CreateModuleSlot(&A);
}
for (const GlobalIFunc &I : TheModule->ifuncs()) {
if (!I.hasName())
CreateModuleSlot(&I);
}
2009-07-30 00:04:47 +02:00
// Add metadata used by named metadata.
for (const NamedMDNode &NMD : TheModule->named_metadata()) {
for (unsigned i = 0, e = NMD.getNumOperands(); i != e; ++i)
CreateMetadataSlot(NMD.getOperand(i));
2009-07-30 00:04:47 +02:00
}
for (const Function &F : *TheModule) {
if (!F.hasName())
// Add all the unnamed functions to the table.
CreateModuleSlot(&F);
IR: Make Metadata::print() reliable and useful Replumb the `AsmWriter` so that `Metadata::print()` is generally useful. (Similarly change `Metadata::printAsOperand()`.) - `SlotTracker` now has a mode where all metadata will be correctly numbered when initializing a `Module`. Normally, `Metadata` only referenced from within `Function`s gets numbered when the `Function` is incorporated. - `Metadata::print()` and `Metadata::printAsOperand()` (and `Metadata::dump()`) now take an optional `Module` argument. When provided, `SlotTracker` is initialized with the new mode, and the numbering will be complete and consistent for all calls to `print()`. - `Value::print()` uses the new `SlotTracker` mode when printing intrinsics with `MDNode` operands, `MetadataAsValue` operands, or the bodies of functions. Thus, metadata numbering will be consistent between calls to `Metadata::print()` and `Value::print()`. - `Metadata::print()` (and `Metadata::dump()`) now print the full definition of `MDNode`s: !5 = !{!6, !"abc", !7} This matches behaviour for `Value::print()`, which includes the name of instructions. - Updated call sites in `Verifier` to call `print()` instead of `printAsOperand()`. All this, so that `Verifier` can print out useful failure messages that involve `Metadata` for PR22777. Note that `Metadata::printAsOperand()` previously took an optional `bool` and `Module` operand. The former was cargo-culted from `Value::printAsOperand()` and wasn't doing anything useful. The latter didn't give consistent results (without the new `SlotTracker` mode). llvm-svn: 232275
2015-03-14 21:19:36 +01:00
if (ShouldInitializeAllMetadata)
processFunctionMetadata(F);
IR: Make Metadata::print() reliable and useful Replumb the `AsmWriter` so that `Metadata::print()` is generally useful. (Similarly change `Metadata::printAsOperand()`.) - `SlotTracker` now has a mode where all metadata will be correctly numbered when initializing a `Module`. Normally, `Metadata` only referenced from within `Function`s gets numbered when the `Function` is incorporated. - `Metadata::print()` and `Metadata::printAsOperand()` (and `Metadata::dump()`) now take an optional `Module` argument. When provided, `SlotTracker` is initialized with the new mode, and the numbering will be complete and consistent for all calls to `print()`. - `Value::print()` uses the new `SlotTracker` mode when printing intrinsics with `MDNode` operands, `MetadataAsValue` operands, or the bodies of functions. Thus, metadata numbering will be consistent between calls to `Metadata::print()` and `Value::print()`. - `Metadata::print()` (and `Metadata::dump()`) now print the full definition of `MDNode`s: !5 = !{!6, !"abc", !7} This matches behaviour for `Value::print()`, which includes the name of instructions. - Updated call sites in `Verifier` to call `print()` instead of `printAsOperand()`. All this, so that `Verifier` can print out useful failure messages that involve `Metadata` for PR22777. Note that `Metadata::printAsOperand()` previously took an optional `bool` and `Module` operand. The former was cargo-culted from `Value::printAsOperand()` and wasn't doing anything useful. The latter didn't give consistent results (without the new `SlotTracker` mode). llvm-svn: 232275
2015-03-14 21:19:36 +01:00
// Add all the function attributes to the table.
// FIXME: Add attributes of other objects?
AttributeSet FnAttrs = F.getAttributes().getFnAttributes();
if (FnAttrs.hasAttributes())
CreateAttributeSetSlot(FnAttrs);
}
ST_DEBUG("end processModule!\n");
}
// Process the arguments, basic blocks, and instructions of a function.
void SlotTracker::processFunction() {
ST_DEBUG("begin processFunction!\n");
fNext = 0;
// Process function metadata if it wasn't hit at the module-level.
if (!ShouldInitializeAllMetadata)
processFunctionMetadata(*TheFunction);
// Add all the function arguments with no names.
for(Function::const_arg_iterator AI = TheFunction->arg_begin(),
AE = TheFunction->arg_end(); AI != AE; ++AI)
if (!AI->hasName())
CreateFunctionSlot(&*AI);
ST_DEBUG("Inserting Instructions:\n");
// Add all of the basic blocks and instructions with no names.
for (auto &BB : *TheFunction) {
if (!BB.hasName())
CreateFunctionSlot(&BB);
2011-09-30 21:48:58 +02:00
for (auto &I : BB) {
if (!I.getType()->isVoidTy() && !I.hasName())
CreateFunctionSlot(&I);
2011-09-30 21:48:58 +02:00
// We allow direct calls to any llvm.foo function here, because the
// target may not be linked into the optimizer.
if (auto CS = ImmutableCallSite(&I)) {
// Add all the call attributes to the table.
AttributeSet Attrs = CS.getAttributes().getFnAttributes();
if (Attrs.hasAttributes())
CreateAttributeSetSlot(Attrs);
}
}
}
FunctionProcessed = true;
ST_DEBUG("end processFunction!\n");
}
void SlotTracker::processGlobalObjectMetadata(const GlobalObject &GO) {
SmallVector<std::pair<unsigned, MDNode *>, 4> MDs;
GO.getAllMetadata(MDs);
for (auto &MD : MDs)
CreateMetadataSlot(MD.second);
}
void SlotTracker::processFunctionMetadata(const Function &F) {
processGlobalObjectMetadata(F);
for (auto &BB : F) {
IR: Make Metadata::print() reliable and useful Replumb the `AsmWriter` so that `Metadata::print()` is generally useful. (Similarly change `Metadata::printAsOperand()`.) - `SlotTracker` now has a mode where all metadata will be correctly numbered when initializing a `Module`. Normally, `Metadata` only referenced from within `Function`s gets numbered when the `Function` is incorporated. - `Metadata::print()` and `Metadata::printAsOperand()` (and `Metadata::dump()`) now take an optional `Module` argument. When provided, `SlotTracker` is initialized with the new mode, and the numbering will be complete and consistent for all calls to `print()`. - `Value::print()` uses the new `SlotTracker` mode when printing intrinsics with `MDNode` operands, `MetadataAsValue` operands, or the bodies of functions. Thus, metadata numbering will be consistent between calls to `Metadata::print()` and `Value::print()`. - `Metadata::print()` (and `Metadata::dump()`) now print the full definition of `MDNode`s: !5 = !{!6, !"abc", !7} This matches behaviour for `Value::print()`, which includes the name of instructions. - Updated call sites in `Verifier` to call `print()` instead of `printAsOperand()`. All this, so that `Verifier` can print out useful failure messages that involve `Metadata` for PR22777. Note that `Metadata::printAsOperand()` previously took an optional `bool` and `Module` operand. The former was cargo-culted from `Value::printAsOperand()` and wasn't doing anything useful. The latter didn't give consistent results (without the new `SlotTracker` mode). llvm-svn: 232275
2015-03-14 21:19:36 +01:00
for (auto &I : BB)
processInstructionMetadata(I);
}
IR: Make Metadata::print() reliable and useful Replumb the `AsmWriter` so that `Metadata::print()` is generally useful. (Similarly change `Metadata::printAsOperand()`.) - `SlotTracker` now has a mode where all metadata will be correctly numbered when initializing a `Module`. Normally, `Metadata` only referenced from within `Function`s gets numbered when the `Function` is incorporated. - `Metadata::print()` and `Metadata::printAsOperand()` (and `Metadata::dump()`) now take an optional `Module` argument. When provided, `SlotTracker` is initialized with the new mode, and the numbering will be complete and consistent for all calls to `print()`. - `Value::print()` uses the new `SlotTracker` mode when printing intrinsics with `MDNode` operands, `MetadataAsValue` operands, or the bodies of functions. Thus, metadata numbering will be consistent between calls to `Metadata::print()` and `Value::print()`. - `Metadata::print()` (and `Metadata::dump()`) now print the full definition of `MDNode`s: !5 = !{!6, !"abc", !7} This matches behaviour for `Value::print()`, which includes the name of instructions. - Updated call sites in `Verifier` to call `print()` instead of `printAsOperand()`. All this, so that `Verifier` can print out useful failure messages that involve `Metadata` for PR22777. Note that `Metadata::printAsOperand()` previously took an optional `bool` and `Module` operand. The former was cargo-culted from `Value::printAsOperand()` and wasn't doing anything useful. The latter didn't give consistent results (without the new `SlotTracker` mode). llvm-svn: 232275
2015-03-14 21:19:36 +01:00
}
void SlotTracker::processInstructionMetadata(const Instruction &I) {
// Process metadata used directly by intrinsics.
if (const CallInst *CI = dyn_cast<CallInst>(&I))
if (Function *F = CI->getCalledFunction())
if (F->isIntrinsic())
for (auto &Op : I.operands())
if (auto *V = dyn_cast_or_null<MetadataAsValue>(Op))
if (MDNode *N = dyn_cast<MDNode>(V->getMetadata()))
CreateMetadataSlot(N);
// Process metadata attached to this instruction.
SmallVector<std::pair<unsigned, MDNode *>, 4> MDs;
I.getAllMetadata(MDs);
for (auto &MD : MDs)
CreateMetadataSlot(MD.second);
}
/// Clean up after incorporating a function. This is the only way to get out of
/// the function incorporation state that affects get*Slot/Create*Slot. Function
/// incorporation state is indicated by TheFunction != 0.
void SlotTracker::purgeFunction() {
ST_DEBUG("begin purgeFunction!\n");
fMap.clear(); // Simply discard the function level map
TheFunction = nullptr;
FunctionProcessed = false;
ST_DEBUG("end purgeFunction!\n");
}
/// getGlobalSlot - Get the slot number of a global value.
int SlotTracker::getGlobalSlot(const GlobalValue *V) {
// Check for uninitialized state and do lazy initialization.
initialize();
// Find the value in the module map
ValueMap::iterator MI = mMap.find(V);
return MI == mMap.end() ? -1 : (int)MI->second;
}
/// getMetadataSlot - Get the slot number of a MDNode.
int SlotTracker::getMetadataSlot(const MDNode *N) {
// Check for uninitialized state and do lazy initialization.
initialize();
// Find the MDNode in the module map
mdn_iterator MI = mdnMap.find(N);
return MI == mdnMap.end() ? -1 : (int)MI->second;
}
/// getLocalSlot - Get the slot number for a value that is local to a function.
int SlotTracker::getLocalSlot(const Value *V) {
assert(!isa<Constant>(V) && "Can't get a constant or global slot with this!");
// Check for uninitialized state and do lazy initialization.
initialize();
ValueMap::iterator FI = fMap.find(V);
return FI == fMap.end() ? -1 : (int)FI->second;
}
int SlotTracker::getAttributeGroupSlot(AttributeSet AS) {
// Check for uninitialized state and do lazy initialization.
initialize();
// Find the AttributeSet in the module map.
as_iterator AI = asMap.find(AS);
return AI == asMap.end() ? -1 : (int)AI->second;
}
/// CreateModuleSlot - Insert the specified GlobalValue* into the slot table.
void SlotTracker::CreateModuleSlot(const GlobalValue *V) {
assert(V && "Can't insert a null Value into SlotTracker!");
assert(!V->getType()->isVoidTy() && "Doesn't need a slot!");
assert(!V->hasName() && "Doesn't need a slot!");
unsigned DestSlot = mNext++;
mMap[V] = DestSlot;
ST_DEBUG(" Inserting value [" << V->getType() << "] = " << V << " slot=" <<
DestSlot << " [");
// G = Global, F = Function, A = Alias, I = IFunc, o = other
ST_DEBUG((isa<GlobalVariable>(V) ? 'G' :
(isa<Function>(V) ? 'F' :
(isa<GlobalAlias>(V) ? 'A' :
(isa<GlobalIFunc>(V) ? 'I' : 'o')))) << "]\n");
}
/// CreateSlot - Create a new slot for the specified value if it has no name.
void SlotTracker::CreateFunctionSlot(const Value *V) {
assert(!V->getType()->isVoidTy() && !V->hasName() && "Doesn't need a slot!");
unsigned DestSlot = fNext++;
fMap[V] = DestSlot;
// G = Global, F = Function, o = other
ST_DEBUG(" Inserting value [" << V->getType() << "] = " << V << " slot=" <<
DestSlot << " [o]\n");
}
/// CreateModuleSlot - Insert the specified MDNode* into the slot table.
void SlotTracker::CreateMetadataSlot(const MDNode *N) {
assert(N && "Can't insert a null Value into SlotTracker!");
// Don't make slots for DIExpressions. We just print them inline everywhere.
if (isa<DIExpression>(N))
return;
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
unsigned DestSlot = mdnNext;
if (!mdnMap.insert(std::make_pair(N, DestSlot)).second)
return;
++mdnNext;
// Recursively add any MDNodes referenced by operands.
for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i)
if (const MDNode *Op = dyn_cast_or_null<MDNode>(N->getOperand(i)))
CreateMetadataSlot(Op);
}
void SlotTracker::CreateAttributeSetSlot(AttributeSet AS) {
assert(AS.hasAttributes() && "Doesn't need a slot!");
as_iterator I = asMap.find(AS);
if (I != asMap.end())
return;
unsigned DestSlot = asNext++;
asMap[AS] = DestSlot;
}
//===----------------------------------------------------------------------===//
// AsmWriter Implementation
//===----------------------------------------------------------------------===//
static void WriteAsOperandInternal(raw_ostream &Out, const Value *V,
TypePrinting *TypePrinter,
SlotTracker *Machine,
const Module *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
static void WriteAsOperandInternal(raw_ostream &Out, const Metadata *MD,
TypePrinting *TypePrinter,
SlotTracker *Machine, const Module *Context,
bool FromValue = false);
static void writeAtomicRMWOperation(raw_ostream &Out,
AtomicRMWInst::BinOp Op) {
switch (Op) {
default: Out << " <unknown operation " << Op << ">"; break;
case AtomicRMWInst::Xchg: Out << " xchg"; break;
case AtomicRMWInst::Add: Out << " add"; break;
case AtomicRMWInst::Sub: Out << " sub"; break;
case AtomicRMWInst::And: Out << " and"; break;
case AtomicRMWInst::Nand: Out << " nand"; break;
case AtomicRMWInst::Or: Out << " or"; break;
case AtomicRMWInst::Xor: Out << " xor"; break;
case AtomicRMWInst::Max: Out << " max"; break;
case AtomicRMWInst::Min: Out << " min"; break;
case AtomicRMWInst::UMax: Out << " umax"; break;
case AtomicRMWInst::UMin: Out << " umin"; break;
}
}
static void WriteOptimizationInfo(raw_ostream &Out, const User *U) {
if (const FPMathOperator *FPO = dyn_cast<const FPMathOperator>(U)) {
[IR] redefine 'UnsafeAlgebra' / 'reassoc' fast-math-flags and add 'trans' fast-math-flag As discussed on llvm-dev: http://lists.llvm.org/pipermail/llvm-dev/2016-November/107104.html and again more recently: http://lists.llvm.org/pipermail/llvm-dev/2017-October/118118.html ...this is a step in cleaning up our fast-math-flags implementation in IR to better match the capabilities of both clang's user-visible flags and the backend's flags for SDNode. As proposed in the above threads, we're replacing the 'UnsafeAlgebra' bit (which had the 'umbrella' meaning that all flags are set) with a new bit that only applies to algebraic reassociation - 'AllowReassoc'. We're also adding a bit to allow approximations for library functions called 'ApproxFunc' (this was initially proposed as 'libm' or similar). ...and we're out of bits. 7 bits ought to be enough for anyone, right? :) FWIW, I did look at getting this out of SubclassOptionalData via SubclassData (spacious 16-bits), but that's apparently already used for other purposes. Also, I don't think we can just add a field to FPMathOperator because Operator is not intended to be instantiated. We'll defer movement of FMF to another day. We keep the 'fast' keyword. I thought about removing that, but seeing IR like this: %f.fast = fadd reassoc nnan ninf nsz arcp contract afn float %op1, %op2 ...made me think we want to keep the shortcut synonym. Finally, this change is binary incompatible with existing IR as seen in the compatibility tests. This statement: "Newer releases can ignore features from older releases, but they cannot miscompile them. For example, if nsw is ever replaced with something else, dropping it would be a valid way to upgrade the IR." ( http://llvm.org/docs/DeveloperPolicy.html#ir-backwards-compatibility ) ...provides the flexibility we want to make this change without requiring a new IR version. Ie, we're not loosening the FP strictness of existing IR. At worst, we will fail to optimize some previously 'fast' code because it's no longer recognized as 'fast'. This should get fixed as we audit/squash all of the uses of 'isFast()'. Note: an inter-dependent clang commit to use the new API name should closely follow commit. Differential Revision: https://reviews.llvm.org/D39304 llvm-svn: 317488
2017-11-06 17:27:15 +01:00
// 'Fast' is an abbreviation for all fast-math-flags.
if (FPO->isFast())
Out << " fast";
else {
[IR] redefine 'UnsafeAlgebra' / 'reassoc' fast-math-flags and add 'trans' fast-math-flag As discussed on llvm-dev: http://lists.llvm.org/pipermail/llvm-dev/2016-November/107104.html and again more recently: http://lists.llvm.org/pipermail/llvm-dev/2017-October/118118.html ...this is a step in cleaning up our fast-math-flags implementation in IR to better match the capabilities of both clang's user-visible flags and the backend's flags for SDNode. As proposed in the above threads, we're replacing the 'UnsafeAlgebra' bit (which had the 'umbrella' meaning that all flags are set) with a new bit that only applies to algebraic reassociation - 'AllowReassoc'. We're also adding a bit to allow approximations for library functions called 'ApproxFunc' (this was initially proposed as 'libm' or similar). ...and we're out of bits. 7 bits ought to be enough for anyone, right? :) FWIW, I did look at getting this out of SubclassOptionalData via SubclassData (spacious 16-bits), but that's apparently already used for other purposes. Also, I don't think we can just add a field to FPMathOperator because Operator is not intended to be instantiated. We'll defer movement of FMF to another day. We keep the 'fast' keyword. I thought about removing that, but seeing IR like this: %f.fast = fadd reassoc nnan ninf nsz arcp contract afn float %op1, %op2 ...made me think we want to keep the shortcut synonym. Finally, this change is binary incompatible with existing IR as seen in the compatibility tests. This statement: "Newer releases can ignore features from older releases, but they cannot miscompile them. For example, if nsw is ever replaced with something else, dropping it would be a valid way to upgrade the IR." ( http://llvm.org/docs/DeveloperPolicy.html#ir-backwards-compatibility ) ...provides the flexibility we want to make this change without requiring a new IR version. Ie, we're not loosening the FP strictness of existing IR. At worst, we will fail to optimize some previously 'fast' code because it's no longer recognized as 'fast'. This should get fixed as we audit/squash all of the uses of 'isFast()'. Note: an inter-dependent clang commit to use the new API name should closely follow commit. Differential Revision: https://reviews.llvm.org/D39304 llvm-svn: 317488
2017-11-06 17:27:15 +01:00
if (FPO->hasAllowReassoc())
Out << " reassoc";
if (FPO->hasNoNaNs())
Out << " nnan";
if (FPO->hasNoInfs())
Out << " ninf";
if (FPO->hasNoSignedZeros())
Out << " nsz";
if (FPO->hasAllowReciprocal())
Out << " arcp";
if (FPO->hasAllowContract())
Out << " contract";
[IR] redefine 'UnsafeAlgebra' / 'reassoc' fast-math-flags and add 'trans' fast-math-flag As discussed on llvm-dev: http://lists.llvm.org/pipermail/llvm-dev/2016-November/107104.html and again more recently: http://lists.llvm.org/pipermail/llvm-dev/2017-October/118118.html ...this is a step in cleaning up our fast-math-flags implementation in IR to better match the capabilities of both clang's user-visible flags and the backend's flags for SDNode. As proposed in the above threads, we're replacing the 'UnsafeAlgebra' bit (which had the 'umbrella' meaning that all flags are set) with a new bit that only applies to algebraic reassociation - 'AllowReassoc'. We're also adding a bit to allow approximations for library functions called 'ApproxFunc' (this was initially proposed as 'libm' or similar). ...and we're out of bits. 7 bits ought to be enough for anyone, right? :) FWIW, I did look at getting this out of SubclassOptionalData via SubclassData (spacious 16-bits), but that's apparently already used for other purposes. Also, I don't think we can just add a field to FPMathOperator because Operator is not intended to be instantiated. We'll defer movement of FMF to another day. We keep the 'fast' keyword. I thought about removing that, but seeing IR like this: %f.fast = fadd reassoc nnan ninf nsz arcp contract afn float %op1, %op2 ...made me think we want to keep the shortcut synonym. Finally, this change is binary incompatible with existing IR as seen in the compatibility tests. This statement: "Newer releases can ignore features from older releases, but they cannot miscompile them. For example, if nsw is ever replaced with something else, dropping it would be a valid way to upgrade the IR." ( http://llvm.org/docs/DeveloperPolicy.html#ir-backwards-compatibility ) ...provides the flexibility we want to make this change without requiring a new IR version. Ie, we're not loosening the FP strictness of existing IR. At worst, we will fail to optimize some previously 'fast' code because it's no longer recognized as 'fast'. This should get fixed as we audit/squash all of the uses of 'isFast()'. Note: an inter-dependent clang commit to use the new API name should closely follow commit. Differential Revision: https://reviews.llvm.org/D39304 llvm-svn: 317488
2017-11-06 17:27:15 +01:00
if (FPO->hasApproxFunc())
Out << " afn";
}
}
if (const OverflowingBinaryOperator *OBO =
dyn_cast<OverflowingBinaryOperator>(U)) {
if (OBO->hasNoUnsignedWrap())
Out << " nuw";
if (OBO->hasNoSignedWrap())
Out << " nsw";
} else if (const PossiblyExactOperator *Div =
dyn_cast<PossiblyExactOperator>(U)) {
if (Div->isExact())
Out << " exact";
} else if (const GEPOperator *GEP = dyn_cast<GEPOperator>(U)) {
if (GEP->isInBounds())
Out << " inbounds";
}
}
static void WriteConstantInternal(raw_ostream &Out, const Constant *CV,
TypePrinting &TypePrinter,
SlotTracker *Machine,
const Module *Context) {
if (const ConstantInt *CI = dyn_cast<ConstantInt>(CV)) {
if (CI->getType()->isIntegerTy(1)) {
Out << (CI->getZExtValue() ? "true" : "false");
return;
}
Out << CI->getValue();
return;
}
if (const ConstantFP *CFP = dyn_cast<ConstantFP>(CV)) {
const APFloat &APF = CFP->getValueAPF();
if (&APF.getSemantics() == &APFloat::IEEEsingle() ||
&APF.getSemantics() == &APFloat::IEEEdouble()) {
// We would like to output the FP constant value in exponential notation,
// but we cannot do this if doing so will lose precision. Check here to
// make sure that we only output it in exponential format if we can parse
// the value back and get the same value.
//
bool ignored;
bool isDouble = &APF.getSemantics() == &APFloat::IEEEdouble();
bool isInf = APF.isInfinity();
bool isNaN = APF.isNaN();
if (!isInf && !isNaN) {
double Val = isDouble ? APF.convertToDouble() : APF.convertToFloat();
SmallString<128> StrVal;
APF.toString(StrVal, 6, 0, false);
// Check to make sure that the stringized number is not some string like
// "Inf" or NaN, that atof will accept, but the lexer will not. Check
// that the string matches the "[-+]?[0-9]" regex.
//
assert(((StrVal[0] >= '0' && StrVal[0] <= '9') ||
((StrVal[0] == '-' || StrVal[0] == '+') &&
(StrVal[1] >= '0' && StrVal[1] <= '9'))) &&
"[-+]?[0-9] regex does not match!");
// Reparse stringized version!
if (APFloat(APFloat::IEEEdouble(), StrVal).convertToDouble() == Val) {
Out << StrVal;
return;
}
}
// Otherwise we could not reparse it to exactly the same value, so we must
// output the string in hexadecimal format! Note that loading and storing
// floating point types changes the bits of NaNs on some hosts, notably
// x86, so we must not use these types.
static_assert(sizeof(double) == sizeof(uint64_t),
"assuming that double is 64 bits!");
APFloat apf = APF;
// Floats are represented in ASCII IR as double, convert.
if (!isDouble)
apf.convert(APFloat::IEEEdouble(), APFloat::rmNearestTiesToEven,
&ignored);
Out << format_hex(apf.bitcastToAPInt().getZExtValue(), 0, /*Upper=*/true);
return;
}
// Either half, or some form of long double.
// These appear as a magic letter identifying the type, then a
// fixed number of hex digits.
Out << "0x";
APInt API = APF.bitcastToAPInt();
if (&APF.getSemantics() == &APFloat::x87DoubleExtended()) {
Out << 'K';
Out << format_hex_no_prefix(API.getHiBits(16).getZExtValue(), 4,
/*Upper=*/true);
Out << format_hex_no_prefix(API.getLoBits(64).getZExtValue(), 16,
/*Upper=*/true);
return;
} else if (&APF.getSemantics() == &APFloat::IEEEquad()) {
Out << 'L';
Out << format_hex_no_prefix(API.getLoBits(64).getZExtValue(), 16,
/*Upper=*/true);
Out << format_hex_no_prefix(API.getHiBits(64).getZExtValue(), 16,
/*Upper=*/true);
} else if (&APF.getSemantics() == &APFloat::PPCDoubleDouble()) {
Out << 'M';
Out << format_hex_no_prefix(API.getLoBits(64).getZExtValue(), 16,
/*Upper=*/true);
Out << format_hex_no_prefix(API.getHiBits(64).getZExtValue(), 16,
/*Upper=*/true);
} else if (&APF.getSemantics() == &APFloat::IEEEhalf()) {
Out << 'H';
Out << format_hex_no_prefix(API.getZExtValue(), 4,
/*Upper=*/true);
} else
llvm_unreachable("Unsupported floating point type");
return;
}
if (isa<ConstantAggregateZero>(CV)) {
Out << "zeroinitializer";
return;
}
2011-09-30 21:48:58 +02:00
if (const BlockAddress *BA = dyn_cast<BlockAddress>(CV)) {
Out << "blockaddress(";
WriteAsOperandInternal(Out, BA->getFunction(), &TypePrinter, Machine,
Context);
Out << ", ";
WriteAsOperandInternal(Out, BA->getBasicBlock(), &TypePrinter, Machine,
Context);
Out << ")";
return;
}
if (const ConstantArray *CA = dyn_cast<ConstantArray>(CV)) {
Type *ETy = CA->getType()->getElementType();
Out << '[';
TypePrinter.print(ETy, Out);
Out << ' ';
WriteAsOperandInternal(Out, CA->getOperand(0),
&TypePrinter, Machine,
Context);
for (unsigned i = 1, e = CA->getNumOperands(); i != e; ++i) {
Out << ", ";
TypePrinter.print(ETy, Out);
Out << ' ';
WriteAsOperandInternal(Out, CA->getOperand(i), &TypePrinter, Machine,
Context);
}
Out << ']';
return;
}
if (const ConstantDataArray *CA = dyn_cast<ConstantDataArray>(CV)) {
// As a special case, print the array as a string if it is an array of
// i8 with ConstantInt values.
if (CA->isString()) {
Out << "c\"";
PrintEscapedString(CA->getAsString(), Out);
Out << '"';
return;
}
Type *ETy = CA->getType()->getElementType();
Out << '[';
TypePrinter.print(ETy, Out);
Out << ' ';
WriteAsOperandInternal(Out, CA->getElementAsConstant(0),
&TypePrinter, Machine,
Context);
for (unsigned i = 1, e = CA->getNumElements(); i != e; ++i) {
Out << ", ";
TypePrinter.print(ETy, Out);
Out << ' ';
WriteAsOperandInternal(Out, CA->getElementAsConstant(i), &TypePrinter,
Machine, Context);
}
Out << ']';
return;
}
if (const ConstantStruct *CS = dyn_cast<ConstantStruct>(CV)) {
if (CS->getType()->isPacked())
Out << '<';
Out << '{';
unsigned N = CS->getNumOperands();
if (N) {
Out << ' ';
TypePrinter.print(CS->getOperand(0)->getType(), Out);
Out << ' ';
WriteAsOperandInternal(Out, CS->getOperand(0), &TypePrinter, Machine,
Context);
for (unsigned i = 1; i < N; i++) {
Out << ", ";
TypePrinter.print(CS->getOperand(i)->getType(), Out);
Out << ' ';
WriteAsOperandInternal(Out, CS->getOperand(i), &TypePrinter, Machine,
Context);
}
Out << ' ';
}
Out << '}';
if (CS->getType()->isPacked())
Out << '>';
return;
}
if (isa<ConstantVector>(CV) || isa<ConstantDataVector>(CV)) {
Type *ETy = CV->getType()->getVectorElementType();
Out << '<';
TypePrinter.print(ETy, Out);
Out << ' ';
WriteAsOperandInternal(Out, CV->getAggregateElement(0U), &TypePrinter,
Machine, Context);
for (unsigned i = 1, e = CV->getType()->getVectorNumElements(); i != e;++i){
Out << ", ";
TypePrinter.print(ETy, Out);
Out << ' ';
WriteAsOperandInternal(Out, CV->getAggregateElement(i), &TypePrinter,
Machine, Context);
}
Out << '>';
return;
}
if (isa<ConstantPointerNull>(CV)) {
Out << "null";
return;
}
if (isa<ConstantTokenNone>(CV)) {
Out << "none";
return;
}
if (isa<UndefValue>(CV)) {
Out << "undef";
return;
}
if (const ConstantExpr *CE = dyn_cast<ConstantExpr>(CV)) {
Out << CE->getOpcodeName();
WriteOptimizationInfo(Out, CE);
if (CE->isCompare())
Out << ' ' << CmpInst::getPredicateName(
static_cast<CmpInst::Predicate>(CE->getPredicate()));
Out << " (";
Optional<unsigned> InRangeOp;
if (const GEPOperator *GEP = dyn_cast<GEPOperator>(CE)) {
TypePrinter.print(GEP->getSourceElementType(), Out);
Out << ", ";
InRangeOp = GEP->getInRangeIndex();
if (InRangeOp)
++*InRangeOp;
}
for (User::const_op_iterator OI=CE->op_begin(); OI != CE->op_end(); ++OI) {
if (InRangeOp && unsigned(OI - CE->op_begin()) == *InRangeOp)
Out << "inrange ";
TypePrinter.print((*OI)->getType(), Out);
Out << ' ';
WriteAsOperandInternal(Out, *OI, &TypePrinter, Machine, Context);
if (OI+1 != CE->op_end())
Out << ", ";
}
if (CE->hasIndices()) {
ArrayRef<unsigned> Indices = CE->getIndices();
for (unsigned i = 0, e = Indices.size(); i != e; ++i)
Out << ", " << Indices[i];
}
if (CE->isCast()) {
Out << " to ";
TypePrinter.print(CE->getType(), Out);
}
Out << ')';
return;
}
Out << "<placeholder or erroneous Constant>";
}
static void writeMDTuple(raw_ostream &Out, const MDTuple *Node,
TypePrinting *TypePrinter, SlotTracker *Machine,
const Module *Context) {
Out << "!{";
for (unsigned mi = 0, me = Node->getNumOperands(); mi != me; ++mi) {
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
const Metadata *MD = Node->getOperand(mi);
if (!MD)
Out << "null";
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
else if (auto *MDV = dyn_cast<ValueAsMetadata>(MD)) {
Value *V = MDV->getValue();
TypePrinter->print(V->getType(), Out);
Out << ' ';
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
WriteAsOperandInternal(Out, V, TypePrinter, Machine, Context);
} else {
WriteAsOperandInternal(Out, MD, TypePrinter, Machine, Context);
}
if (mi + 1 != me)
Out << ", ";
}
2011-09-30 21:48:58 +02:00
Out << "}";
}
namespace {
struct FieldSeparator {
bool Skip = true;
const char *Sep;
FieldSeparator(const char *Sep = ", ") : Sep(Sep) {}
};
raw_ostream &operator<<(raw_ostream &OS, FieldSeparator &FS) {
if (FS.Skip) {
FS.Skip = false;
return OS;
}
return OS << FS.Sep;
}
struct MDFieldPrinter {
raw_ostream &Out;
FieldSeparator FS;
TypePrinting *TypePrinter = nullptr;
SlotTracker *Machine = nullptr;
const Module *Context = nullptr;
explicit MDFieldPrinter(raw_ostream &Out) : Out(Out) {}
MDFieldPrinter(raw_ostream &Out, TypePrinting *TypePrinter,
SlotTracker *Machine, const Module *Context)
: Out(Out), TypePrinter(TypePrinter), Machine(Machine), Context(Context) {
}
void printTag(const DINode *N);
void printMacinfoType(const DIMacroNode *N);
void printChecksum(const DIFile::ChecksumInfo<StringRef> &N);
void printString(StringRef Name, StringRef Value,
bool ShouldSkipEmpty = true);
void printMetadata(StringRef Name, const Metadata *MD,
bool ShouldSkipNull = true);
template <class IntTy>
void printInt(StringRef Name, IntTy Int, bool ShouldSkipZero = true);
void printBool(StringRef Name, bool Value, Optional<bool> Default = None);
void printDIFlags(StringRef Name, DINode::DIFlags Flags);
template <class IntTy, class Stringifier>
void printDwarfEnum(StringRef Name, IntTy Value, Stringifier toString,
bool ShouldSkipZero = true);
void printEmissionKind(StringRef Name, DICompileUnit::DebugEmissionKind EK);
};
} // end anonymous namespace
void MDFieldPrinter::printTag(const DINode *N) {
Out << FS << "tag: ";
auto Tag = dwarf::TagString(N->getTag());
if (!Tag.empty())
Out << Tag;
else
Out << N->getTag();
}
void MDFieldPrinter::printMacinfoType(const DIMacroNode *N) {
Out << FS << "type: ";
auto Type = dwarf::MacinfoString(N->getMacinfoType());
if (!Type.empty())
Out << Type;
else
Out << N->getMacinfoType();
}
void MDFieldPrinter::printChecksum(
const DIFile::ChecksumInfo<StringRef> &Checksum) {
Out << FS << "checksumkind: " << Checksum.getKindAsString();
printString("checksum", Checksum.Value, /* ShouldSkipEmpty */ false);
}
void MDFieldPrinter::printString(StringRef Name, StringRef Value,
bool ShouldSkipEmpty) {
if (ShouldSkipEmpty && Value.empty())
return;
Out << FS << Name << ": \"";
PrintEscapedString(Value, Out);
Out << "\"";
}
static void writeMetadataAsOperand(raw_ostream &Out, const Metadata *MD,
TypePrinting *TypePrinter,
SlotTracker *Machine,
const Module *Context) {
if (!MD) {
Out << "null";
return;
}
WriteAsOperandInternal(Out, MD, TypePrinter, Machine, Context);
}
void MDFieldPrinter::printMetadata(StringRef Name, const Metadata *MD,
bool ShouldSkipNull) {
if (ShouldSkipNull && !MD)
return;
Out << FS << Name << ": ";
writeMetadataAsOperand(Out, MD, TypePrinter, Machine, Context);
}
template <class IntTy>
void MDFieldPrinter::printInt(StringRef Name, IntTy Int, bool ShouldSkipZero) {
if (ShouldSkipZero && !Int)
return;
Out << FS << Name << ": " << Int;
}
void MDFieldPrinter::printBool(StringRef Name, bool Value,
Optional<bool> Default) {
if (Default && Value == *Default)
return;
Out << FS << Name << ": " << (Value ? "true" : "false");
}
void MDFieldPrinter::printDIFlags(StringRef Name, DINode::DIFlags Flags) {
if (!Flags)
return;
Out << FS << Name << ": ";
SmallVector<DINode::DIFlags, 8> SplitFlags;
auto Extra = DINode::splitFlags(Flags, SplitFlags);
FieldSeparator FlagsFS(" | ");
for (auto F : SplitFlags) {
auto StringF = DINode::getFlagString(F);
assert(!StringF.empty() && "Expected valid flag");
Out << FlagsFS << StringF;
}
if (Extra || SplitFlags.empty())
Out << FlagsFS << Extra;
}
void MDFieldPrinter::printEmissionKind(StringRef Name,
DICompileUnit::DebugEmissionKind EK) {
Out << FS << Name << ": " << DICompileUnit::EmissionKindString(EK);
}
template <class IntTy, class Stringifier>
void MDFieldPrinter::printDwarfEnum(StringRef Name, IntTy Value,
Stringifier toString, bool ShouldSkipZero) {
if (!Value)
return;
Out << FS << Name << ": ";
auto S = toString(Value);
if (!S.empty())
Out << S;
else
Out << Value;
}
static void writeGenericDINode(raw_ostream &Out, const GenericDINode *N,
TypePrinting *TypePrinter, SlotTracker *Machine,
const Module *Context) {
Out << "!GenericDINode(";
MDFieldPrinter Printer(Out, TypePrinter, Machine, Context);
Printer.printTag(N);
Printer.printString("header", N->getHeader());
if (N->getNumDwarfOperands()) {
Out << Printer.FS << "operands: {";
FieldSeparator IFS;
for (auto &I : N->dwarf_operands()) {
Out << IFS;
writeMetadataAsOperand(Out, I, TypePrinter, Machine, Context);
}
Out << "}";
}
Out << ")";
}
static void writeDILocation(raw_ostream &Out, const DILocation *DL,
TypePrinting *TypePrinter, SlotTracker *Machine,
const Module *Context) {
Out << "!DILocation(";
MDFieldPrinter Printer(Out, TypePrinter, Machine, Context);
// Always output the line, since 0 is a relevant and important value for it.
Printer.printInt("line", DL->getLine(), /* ShouldSkipZero */ false);
Printer.printInt("column", DL->getColumn());
Printer.printMetadata("scope", DL->getRawScope(), /* ShouldSkipNull */ false);
Printer.printMetadata("inlinedAt", DL->getRawInlinedAt());
Out << ")";
}
static void writeDISubrange(raw_ostream &Out, const DISubrange *N,
TypePrinting *TypePrinter, SlotTracker *Machine,
const Module *Context) {
Out << "!DISubrange(";
MDFieldPrinter Printer(Out, TypePrinter, Machine, Context);
if (auto *CE = N->getCount().dyn_cast<ConstantInt*>())
Printer.printInt("count", CE->getSExtValue(), /* ShouldSkipZero */ false);
else
Printer.printMetadata("count", N->getCount().dyn_cast<DIVariable*>(),
/*ShouldSkipNull */ false);
Printer.printInt("lowerBound", N->getLowerBound());
Out << ")";
}
static void writeDIEnumerator(raw_ostream &Out, const DIEnumerator *N,
TypePrinting *, SlotTracker *, const Module *) {
Out << "!DIEnumerator(";
MDFieldPrinter Printer(Out);
Printer.printString("name", N->getName(), /* ShouldSkipEmpty */ false);
if (N->isUnsigned()) {
auto Value = static_cast<uint64_t>(N->getValue());
Printer.printInt("value", Value, /* ShouldSkipZero */ false);
Printer.printBool("isUnsigned", true);
} else {
Printer.printInt("value", N->getValue(), /* ShouldSkipZero */ false);
}
Out << ")";
}
static void writeDIBasicType(raw_ostream &Out, const DIBasicType *N,
TypePrinting *, SlotTracker *, const Module *) {
Out << "!DIBasicType(";
MDFieldPrinter Printer(Out);
if (N->getTag() != dwarf::DW_TAG_base_type)
Printer.printTag(N);
Printer.printString("name", N->getName());
Printer.printInt("size", N->getSizeInBits());
Printer.printInt("align", N->getAlignInBits());
Printer.printDwarfEnum("encoding", N->getEncoding(),
dwarf::AttributeEncodingString);
Out << ")";
}
static void writeDIDerivedType(raw_ostream &Out, const DIDerivedType *N,
TypePrinting *TypePrinter, SlotTracker *Machine,
const Module *Context) {
Out << "!DIDerivedType(";
MDFieldPrinter Printer(Out, TypePrinter, Machine, Context);
Printer.printTag(N);
Printer.printString("name", N->getName());
Printer.printMetadata("scope", N->getRawScope());
Printer.printMetadata("file", N->getRawFile());
Printer.printInt("line", N->getLine());
Printer.printMetadata("baseType", N->getRawBaseType(),
/* ShouldSkipNull */ false);
Printer.printInt("size", N->getSizeInBits());
Printer.printInt("align", N->getAlignInBits());
Printer.printInt("offset", N->getOffsetInBits());
Printer.printDIFlags("flags", N->getFlags());
Printer.printMetadata("extraData", N->getRawExtraData());
if (const auto &DWARFAddressSpace = N->getDWARFAddressSpace())
Printer.printInt("dwarfAddressSpace", *DWARFAddressSpace,
/* ShouldSkipZero */ false);
Out << ")";
}
static void writeDICompositeType(raw_ostream &Out, const DICompositeType *N,
TypePrinting *TypePrinter,
SlotTracker *Machine, const Module *Context) {
Out << "!DICompositeType(";
MDFieldPrinter Printer(Out, TypePrinter, Machine, Context);
Printer.printTag(N);
Printer.printString("name", N->getName());
Printer.printMetadata("scope", N->getRawScope());
Printer.printMetadata("file", N->getRawFile());
Printer.printInt("line", N->getLine());
Printer.printMetadata("baseType", N->getRawBaseType());
Printer.printInt("size", N->getSizeInBits());
Printer.printInt("align", N->getAlignInBits());
Printer.printInt("offset", N->getOffsetInBits());
Printer.printDIFlags("flags", N->getFlags());
Printer.printMetadata("elements", N->getRawElements());
Printer.printDwarfEnum("runtimeLang", N->getRuntimeLang(),
dwarf::LanguageString);
Printer.printMetadata("vtableHolder", N->getRawVTableHolder());
Printer.printMetadata("templateParams", N->getRawTemplateParams());
Printer.printString("identifier", N->getIdentifier());
Printer.printMetadata("discriminator", N->getRawDiscriminator());
Out << ")";
}
static void writeDISubroutineType(raw_ostream &Out, const DISubroutineType *N,
TypePrinting *TypePrinter,
SlotTracker *Machine, const Module *Context) {
Out << "!DISubroutineType(";
MDFieldPrinter Printer(Out, TypePrinter, Machine, Context);
Printer.printDIFlags("flags", N->getFlags());
Printer.printDwarfEnum("cc", N->getCC(), dwarf::ConventionString);
Printer.printMetadata("types", N->getRawTypeArray(),
/* ShouldSkipNull */ false);
Out << ")";
}
static void writeDIFile(raw_ostream &Out, const DIFile *N, TypePrinting *,
SlotTracker *, const Module *) {
Out << "!DIFile(";
MDFieldPrinter Printer(Out);
Printer.printString("filename", N->getFilename(),
/* ShouldSkipEmpty */ false);
Printer.printString("directory", N->getDirectory(),
/* ShouldSkipEmpty */ false);
// Print all values for checksum together, or not at all.
if (N->getChecksum())
Printer.printChecksum(*N->getChecksum());
Printer.printString("source", N->getSource().getValueOr(StringRef()),
/* ShouldSkipEmpty */ true);
Out << ")";
}
static void writeDICompileUnit(raw_ostream &Out, const DICompileUnit *N,
TypePrinting *TypePrinter, SlotTracker *Machine,
const Module *Context) {
Out << "!DICompileUnit(";
MDFieldPrinter Printer(Out, TypePrinter, Machine, Context);
Printer.printDwarfEnum("language", N->getSourceLanguage(),
dwarf::LanguageString, /* ShouldSkipZero */ false);
Printer.printMetadata("file", N->getRawFile(), /* ShouldSkipNull */ false);
Printer.printString("producer", N->getProducer());
Printer.printBool("isOptimized", N->isOptimized());
Printer.printString("flags", N->getFlags());
Printer.printInt("runtimeVersion", N->getRuntimeVersion(),
/* ShouldSkipZero */ false);
Printer.printString("splitDebugFilename", N->getSplitDebugFilename());
Printer.printEmissionKind("emissionKind", N->getEmissionKind());
Printer.printMetadata("enums", N->getRawEnumTypes());
Printer.printMetadata("retainedTypes", N->getRawRetainedTypes());
Printer.printMetadata("globals", N->getRawGlobalVariables());
Printer.printMetadata("imports", N->getRawImportedEntities());
Printer.printMetadata("macros", N->getRawMacros());
Printer.printInt("dwoId", N->getDWOId());
Printer.printBool("splitDebugInlining", N->getSplitDebugInlining(), true);
Printer.printBool("debugInfoForProfiling", N->getDebugInfoForProfiling(),
false);
Printer.printBool("gnuPubnames", N->getGnuPubnames(), false);
Out << ")";
}
static void writeDISubprogram(raw_ostream &Out, const DISubprogram *N,
TypePrinting *TypePrinter, SlotTracker *Machine,
const Module *Context) {
Out << "!DISubprogram(";
MDFieldPrinter Printer(Out, TypePrinter, Machine, Context);
Printer.printString("name", N->getName());
Printer.printString("linkageName", N->getLinkageName());
Printer.printMetadata("scope", N->getRawScope(), /* ShouldSkipNull */ false);
Printer.printMetadata("file", N->getRawFile());
Printer.printInt("line", N->getLine());
Printer.printMetadata("type", N->getRawType());
Printer.printBool("isLocal", N->isLocalToUnit());
Printer.printBool("isDefinition", N->isDefinition());
Printer.printInt("scopeLine", N->getScopeLine());
Printer.printMetadata("containingType", N->getRawContainingType());
Printer.printDwarfEnum("virtuality", N->getVirtuality(),
dwarf::VirtualityString);
if (N->getVirtuality() != dwarf::DW_VIRTUALITY_none ||
N->getVirtualIndex() != 0)
Printer.printInt("virtualIndex", N->getVirtualIndex(), false);
Printer.printInt("thisAdjustment", N->getThisAdjustment());
Printer.printDIFlags("flags", N->getFlags());
Printer.printBool("isOptimized", N->isOptimized());
Printer.printMetadata("unit", N->getRawUnit());
Printer.printMetadata("templateParams", N->getRawTemplateParams());
Printer.printMetadata("declaration", N->getRawDeclaration());
Printer.printMetadata("variables", N->getRawVariables());
Printer.printMetadata("thrownTypes", N->getRawThrownTypes());
Out << ")";
}
static void writeDILexicalBlock(raw_ostream &Out, const DILexicalBlock *N,
TypePrinting *TypePrinter, SlotTracker *Machine,
const Module *Context) {
Out << "!DILexicalBlock(";
MDFieldPrinter Printer(Out, TypePrinter, Machine, Context);
Printer.printMetadata("scope", N->getRawScope(), /* ShouldSkipNull */ false);
Printer.printMetadata("file", N->getRawFile());
Printer.printInt("line", N->getLine());
Printer.printInt("column", N->getColumn());
Out << ")";
}
static void writeDILexicalBlockFile(raw_ostream &Out,
const DILexicalBlockFile *N,
TypePrinting *TypePrinter,
SlotTracker *Machine,
const Module *Context) {
Out << "!DILexicalBlockFile(";
MDFieldPrinter Printer(Out, TypePrinter, Machine, Context);
Printer.printMetadata("scope", N->getRawScope(), /* ShouldSkipNull */ false);
Printer.printMetadata("file", N->getRawFile());
Printer.printInt("discriminator", N->getDiscriminator(),
/* ShouldSkipZero */ false);
Out << ")";
}
static void writeDINamespace(raw_ostream &Out, const DINamespace *N,
TypePrinting *TypePrinter, SlotTracker *Machine,
const Module *Context) {
Out << "!DINamespace(";
MDFieldPrinter Printer(Out, TypePrinter, Machine, Context);
Printer.printString("name", N->getName());
Printer.printMetadata("scope", N->getRawScope(), /* ShouldSkipNull */ false);
Printer.printBool("exportSymbols", N->getExportSymbols(), false);
Out << ")";
}
static void writeDIMacro(raw_ostream &Out, const DIMacro *N,
TypePrinting *TypePrinter, SlotTracker *Machine,
const Module *Context) {
Out << "!DIMacro(";
MDFieldPrinter Printer(Out, TypePrinter, Machine, Context);
Printer.printMacinfoType(N);
Printer.printInt("line", N->getLine());
Printer.printString("name", N->getName());
Printer.printString("value", N->getValue());
Out << ")";
}
static void writeDIMacroFile(raw_ostream &Out, const DIMacroFile *N,
TypePrinting *TypePrinter, SlotTracker *Machine,
const Module *Context) {
Out << "!DIMacroFile(";
MDFieldPrinter Printer(Out, TypePrinter, Machine, Context);
Printer.printInt("line", N->getLine());
Printer.printMetadata("file", N->getRawFile(), /* ShouldSkipNull */ false);
Printer.printMetadata("nodes", N->getRawElements());
Out << ")";
}
static void writeDIModule(raw_ostream &Out, const DIModule *N,
TypePrinting *TypePrinter, SlotTracker *Machine,
const Module *Context) {
Out << "!DIModule(";
MDFieldPrinter Printer(Out, TypePrinter, Machine, Context);
Printer.printMetadata("scope", N->getRawScope(), /* ShouldSkipNull */ false);
Printer.printString("name", N->getName());
Printer.printString("configMacros", N->getConfigurationMacros());
Printer.printString("includePath", N->getIncludePath());
Printer.printString("isysroot", N->getISysRoot());
Out << ")";
}
static void writeDITemplateTypeParameter(raw_ostream &Out,
const DITemplateTypeParameter *N,
TypePrinting *TypePrinter,
SlotTracker *Machine,
const Module *Context) {
Out << "!DITemplateTypeParameter(";
MDFieldPrinter Printer(Out, TypePrinter, Machine, Context);
Printer.printString("name", N->getName());
Printer.printMetadata("type", N->getRawType(), /* ShouldSkipNull */ false);
Out << ")";
}
static void writeDITemplateValueParameter(raw_ostream &Out,
const DITemplateValueParameter *N,
TypePrinting *TypePrinter,
SlotTracker *Machine,
const Module *Context) {
Out << "!DITemplateValueParameter(";
MDFieldPrinter Printer(Out, TypePrinter, Machine, Context);
if (N->getTag() != dwarf::DW_TAG_template_value_parameter)
Printer.printTag(N);
Printer.printString("name", N->getName());
Printer.printMetadata("type", N->getRawType());
Printer.printMetadata("value", N->getValue(), /* ShouldSkipNull */ false);
Out << ")";
}
static void writeDIGlobalVariable(raw_ostream &Out, const DIGlobalVariable *N,
TypePrinting *TypePrinter,
SlotTracker *Machine, const Module *Context) {
Out << "!DIGlobalVariable(";
MDFieldPrinter Printer(Out, TypePrinter, Machine, Context);
Printer.printString("name", N->getName());
Printer.printString("linkageName", N->getLinkageName());
Printer.printMetadata("scope", N->getRawScope(), /* ShouldSkipNull */ false);
Printer.printMetadata("file", N->getRawFile());
Printer.printInt("line", N->getLine());
Printer.printMetadata("type", N->getRawType());
Printer.printBool("isLocal", N->isLocalToUnit());
Printer.printBool("isDefinition", N->isDefinition());
Printer.printMetadata("declaration", N->getRawStaticDataMemberDeclaration());
Printer.printInt("align", N->getAlignInBits());
Out << ")";
}
static void writeDILocalVariable(raw_ostream &Out, const DILocalVariable *N,
TypePrinting *TypePrinter,
SlotTracker *Machine, const Module *Context) {
Out << "!DILocalVariable(";
MDFieldPrinter Printer(Out, TypePrinter, Machine, Context);
Printer.printString("name", N->getName());
Printer.printInt("arg", N->getArg());
Printer.printMetadata("scope", N->getRawScope(), /* ShouldSkipNull */ false);
Printer.printMetadata("file", N->getRawFile());
Printer.printInt("line", N->getLine());
Printer.printMetadata("type", N->getRawType());
Printer.printDIFlags("flags", N->getFlags());
Printer.printInt("align", N->getAlignInBits());
Out << ")";
}
static void writeDIExpression(raw_ostream &Out, const DIExpression *N,
TypePrinting *TypePrinter, SlotTracker *Machine,
const Module *Context) {
Out << "!DIExpression(";
FieldSeparator FS;
if (N->isValid()) {
for (auto I = N->expr_op_begin(), E = N->expr_op_end(); I != E; ++I) {
auto OpStr = dwarf::OperationEncodingString(I->getOp());
assert(!OpStr.empty() && "Expected valid opcode");
Out << FS << OpStr;
for (unsigned A = 0, AE = I->getNumArgs(); A != AE; ++A)
Out << FS << I->getArg(A);
}
} else {
for (const auto &I : N->getElements())
Out << FS << I;
}
Out << ")";
}
static void writeDIGlobalVariableExpression(raw_ostream &Out,
const DIGlobalVariableExpression *N,
TypePrinting *TypePrinter,
SlotTracker *Machine,
const Module *Context) {
Out << "!DIGlobalVariableExpression(";
MDFieldPrinter Printer(Out, TypePrinter, Machine, Context);
Printer.printMetadata("var", N->getVariable());
Printer.printMetadata("expr", N->getExpression());
Out << ")";
}
static void writeDIObjCProperty(raw_ostream &Out, const DIObjCProperty *N,
TypePrinting *TypePrinter, SlotTracker *Machine,
const Module *Context) {
Out << "!DIObjCProperty(";
MDFieldPrinter Printer(Out, TypePrinter, Machine, Context);
Printer.printString("name", N->getName());
Printer.printMetadata("file", N->getRawFile());
Printer.printInt("line", N->getLine());
Printer.printString("setter", N->getSetterName());
Printer.printString("getter", N->getGetterName());
Printer.printInt("attributes", N->getAttributes());
Printer.printMetadata("type", N->getRawType());
Out << ")";
}
static void writeDIImportedEntity(raw_ostream &Out, const DIImportedEntity *N,
TypePrinting *TypePrinter,
SlotTracker *Machine, const Module *Context) {
Out << "!DIImportedEntity(";
MDFieldPrinter Printer(Out, TypePrinter, Machine, Context);
Printer.printTag(N);
Printer.printString("name", N->getName());
Printer.printMetadata("scope", N->getRawScope(), /* ShouldSkipNull */ false);
Printer.printMetadata("entity", N->getRawEntity());
Printer.printMetadata("file", N->getRawFile());
Printer.printInt("line", N->getLine());
Out << ")";
}
static void WriteMDNodeBodyInternal(raw_ostream &Out, const MDNode *Node,
TypePrinting *TypePrinter,
SlotTracker *Machine,
const Module *Context) {
if (Node->isDistinct())
Out << "distinct ";
else if (Node->isTemporary())
Out << "<temporary!> "; // Handle broken code.
switch (Node->getMetadataID()) {
default:
llvm_unreachable("Expected uniquable MDNode");
#define HANDLE_MDNODE_LEAF(CLASS) \
case Metadata::CLASS##Kind: \
write##CLASS(Out, cast<CLASS>(Node), TypePrinter, Machine, Context); \
break;
#include "llvm/IR/Metadata.def"
}
}
// Full implementation of printing a Value as an operand with support for
// TypePrinting, etc.
static void WriteAsOperandInternal(raw_ostream &Out, const Value *V,
TypePrinting *TypePrinter,
SlotTracker *Machine,
const Module *Context) {
if (V->hasName()) {
PrintLLVMName(Out, V);
return;
}
const Constant *CV = dyn_cast<Constant>(V);
if (CV && !isa<GlobalValue>(CV)) {
assert(TypePrinter && "Constants require TypePrinting!");
WriteConstantInternal(Out, CV, *TypePrinter, Machine, Context);
return;
}
if (const InlineAsm *IA = dyn_cast<InlineAsm>(V)) {
Out << "asm ";
if (IA->hasSideEffects())
Out << "sideeffect ";
if (IA->isAlignStack())
Out << "alignstack ";
// We don't emit the AD_ATT dialect as it's the assumed default.
if (IA->getDialect() == InlineAsm::AD_Intel)
Out << "inteldialect ";
Out << '"';
PrintEscapedString(IA->getAsmString(), Out);
Out << "\", \"";
PrintEscapedString(IA->getConstraintString(), Out);
Out << '"';
return;
}
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
if (auto *MD = dyn_cast<MetadataAsValue>(V)) {
WriteAsOperandInternal(Out, MD->getMetadata(), TypePrinter, Machine,
Context, /* FromValue */ true);
return;
}
char Prefix = '%';
int Slot;
// If we have a SlotTracker, use it.
if (Machine) {
if (const GlobalValue *GV = dyn_cast<GlobalValue>(V)) {
Slot = Machine->getGlobalSlot(GV);
Prefix = '@';
} else {
Slot = Machine->getLocalSlot(V);
2011-09-30 21:48:58 +02:00
// If the local value didn't succeed, then we may be referring to a value
// from a different function. Translate it, as this can happen when using
// address of blocks.
if (Slot == -1)
if ((Machine = createSlotTracker(V))) {
Slot = Machine->getLocalSlot(V);
delete Machine;
}
}
} else if ((Machine = createSlotTracker(V))) {
// Otherwise, create one to get the # and then destroy it.
if (const GlobalValue *GV = dyn_cast<GlobalValue>(V)) {
Slot = Machine->getGlobalSlot(GV);
Prefix = '@';
} else {
Slot = Machine->getLocalSlot(V);
}
delete Machine;
Machine = nullptr;
} else {
Slot = -1;
}
if (Slot != -1)
Out << Prefix << Slot;
else
Out << "<badref>";
}
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
static void WriteAsOperandInternal(raw_ostream &Out, const Metadata *MD,
TypePrinting *TypePrinter,
SlotTracker *Machine, const Module *Context,
bool FromValue) {
// Write DIExpressions inline when used as a value. Improves readability of
// debug info intrinsics.
if (const DIExpression *Expr = dyn_cast<DIExpression>(MD)) {
writeDIExpression(Out, Expr, TypePrinter, Machine, Context);
return;
}
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
if (const MDNode *N = dyn_cast<MDNode>(MD)) {
std::unique_ptr<SlotTracker> MachineStorage;
if (!Machine) {
MachineStorage = make_unique<SlotTracker>(Context);
Machine = MachineStorage.get();
}
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
int Slot = Machine->getMetadataSlot(N);
if (Slot == -1)
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
// Give the pointer value instead of "badref", since this comes up all
// the time when debugging.
Out << "<" << 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
else
Out << '!' << Slot;
return;
}
if (const MDString *MDS = dyn_cast<MDString>(MD)) {
Out << "!\"";
PrintEscapedString(MDS->getString(), Out);
Out << '"';
return;
}
auto *V = cast<ValueAsMetadata>(MD);
assert(TypePrinter && "TypePrinter required for metadata values");
assert((FromValue || !isa<LocalAsMetadata>(V)) &&
"Unexpected function-local metadata outside of value argument");
TypePrinter->print(V->getValue()->getType(), Out);
Out << ' ';
WriteAsOperandInternal(Out, V->getValue(), TypePrinter, Machine, Context);
}
namespace {
class AssemblyWriter {
formatted_raw_ostream &Out;
const Module *TheModule;
std::unique_ptr<SlotTracker> SlotTrackerStorage;
SlotTracker &Machine;
TypePrinting TypePrinter;
AssemblyAnnotationWriter *AnnotationWriter;
SetVector<const Comdat *> Comdats;
bool IsForDebug;
bool ShouldPreserveUseListOrder;
UseListOrderStack UseListOrders;
SmallVector<StringRef, 8> MDNames;
/// Synchronization scope names registered with LLVMContext.
SmallVector<StringRef, 8> SSNs;
public:
/// Construct an AssemblyWriter with an external SlotTracker
AssemblyWriter(formatted_raw_ostream &o, SlotTracker &Mac, const Module *M,
AssemblyAnnotationWriter *AAW, bool IsForDebug,
bool ShouldPreserveUseListOrder = false);
void printMDNodeBody(const MDNode *MD);
void printNamedMDNode(const NamedMDNode *NMD);
void printModule(const Module *M);
void writeOperand(const Value *Op, bool PrintType);
void writeParamOperand(const Value *Operand, AttributeSet Attrs);
void writeOperandBundles(ImmutableCallSite CS);
void writeSyncScope(const LLVMContext &Context,
SyncScope::ID SSID);
void writeAtomic(const LLVMContext &Context,
AtomicOrdering Ordering,
SyncScope::ID SSID);
void writeAtomicCmpXchg(const LLVMContext &Context,
AtomicOrdering SuccessOrdering,
AtomicOrdering FailureOrdering,
SyncScope::ID SSID);
void writeAllMDNodes();
void writeMDNode(unsigned Slot, const MDNode *Node);
void writeAllAttributeGroups();
void printTypeIdentities();
void printGlobal(const GlobalVariable *GV);
void printIndirectSymbol(const GlobalIndirectSymbol *GIS);
void printComdat(const Comdat *C);
void printFunction(const Function *F);
void printArgument(const Argument *FA, AttributeSet Attrs);
void printBasicBlock(const BasicBlock *BB);
void printInstructionLine(const Instruction &I);
void printInstruction(const Instruction &I);
void printUseListOrder(const UseListOrder &Order);
void printUseLists(const Function *F);
private:
/// Print out metadata attachments.
void printMetadataAttachments(
const SmallVectorImpl<std::pair<unsigned, MDNode *>> &MDs,
StringRef Separator);
// printInfoComment - Print a little comment after the instruction indicating
// which slot it occupies.
void printInfoComment(const Value &V);
// printGCRelocateComment - print comment after call to the gc.relocate
// intrinsic indicating base and derived pointer names.
void printGCRelocateComment(const GCRelocateInst &Relocate);
};
} // end anonymous namespace
AssemblyWriter::AssemblyWriter(formatted_raw_ostream &o, SlotTracker &Mac,
const Module *M, AssemblyAnnotationWriter *AAW,
bool IsForDebug, bool ShouldPreserveUseListOrder)
[MIR] Making MIR Printing, opt -dot-cfg, and -debug printing faster Value::printAsOperand has been scanning the entire module just to print a single value as an operand, regardless being asked to print a type or not at all, and regardless really needing to scan the module to print a type. It made some of the users of the method exceptionally slow on large IR-modules (or large MIR-files with large IR-modules embedded). This patch defers scanning a module looking for struct types, mostly numbered struct types, as much as possible, speeding up those users w/o changing any APIs at all. See speedup examples below: Release Build: # 83 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 133 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output Release + Asserts Build: # 95 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 146 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output # 1096 seconds -> 553 seconds time ./bin/llc -debug-only=isel -fast-isel=false -stop-after=isel \ sqlite3.O0.ll -o /dev/null 2> err where sqlite3.O0.ll is non-optimized IR produced from sqlite-amalgamation (http://sqlite.org/download.html), which is entire SQLite3 implementation in a single C-file. Benchmarked on 4-cores / 8 threads PCI-E SSD iMac running macOS Reviewers: dexonsmith, bkramer, void, chandlerc, aditya_nandakumar, dsanders, qcolombet, Reviewed By: bogner Subscribers: thegameg, llvm-commits Differential Revision: https://reviews.llvm.org/D44132 llvm-svn: 328246
2018-03-22 22:29:07 +01:00
: Out(o), TheModule(M), Machine(Mac), TypePrinter(M), AnnotationWriter(AAW),
IsForDebug(IsForDebug),
ShouldPreserveUseListOrder(ShouldPreserveUseListOrder) {
if (!TheModule)
return;
for (const GlobalObject &GO : TheModule->global_objects())
if (const Comdat *C = GO.getComdat())
Comdats.insert(C);
}
2011-09-30 21:48:58 +02:00
void AssemblyWriter::writeOperand(const Value *Operand, bool PrintType) {
if (!Operand) {
Out << "<null operand!>";
return;
}
if (PrintType) {
TypePrinter.print(Operand->getType(), Out);
Out << ' ';
}
WriteAsOperandInternal(Out, Operand, &TypePrinter, &Machine, TheModule);
}
void AssemblyWriter::writeSyncScope(const LLVMContext &Context,
SyncScope::ID SSID) {
switch (SSID) {
case SyncScope::System: {
break;
}
default: {
if (SSNs.empty())
Context.getSyncScopeNames(SSNs);
Out << " syncscope(\"";
PrintEscapedString(SSNs[SSID], Out);
Out << "\")";
break;
}
}
}
void AssemblyWriter::writeAtomic(const LLVMContext &Context,
AtomicOrdering Ordering,
SyncScope::ID SSID) {
if (Ordering == AtomicOrdering::NotAtomic)
return;
writeSyncScope(Context, SSID);
Out << " " << toIRString(Ordering);
}
void AssemblyWriter::writeAtomicCmpXchg(const LLVMContext &Context,
AtomicOrdering SuccessOrdering,
AtomicOrdering FailureOrdering,
SyncScope::ID SSID) {
assert(SuccessOrdering != AtomicOrdering::NotAtomic &&
FailureOrdering != AtomicOrdering::NotAtomic);
writeSyncScope(Context, SSID);
Out << " " << toIRString(SuccessOrdering);
Out << " " << toIRString(FailureOrdering);
}
void AssemblyWriter::writeParamOperand(const Value *Operand,
AttributeSet Attrs) {
if (!Operand) {
Out << "<null operand!>";
return;
}
// Print the type
TypePrinter.print(Operand->getType(), Out);
// Print parameter attributes list
if (Attrs.hasAttributes())
Out << ' ' << Attrs.getAsString();
Out << ' ';
// Print the operand
WriteAsOperandInternal(Out, Operand, &TypePrinter, &Machine, TheModule);
}
void AssemblyWriter::writeOperandBundles(ImmutableCallSite CS) {
if (!CS.hasOperandBundles())
return;
Out << " [ ";
bool FirstBundle = true;
for (unsigned i = 0, e = CS.getNumOperandBundles(); i != e; ++i) {
OperandBundleUse BU = CS.getOperandBundleAt(i);
if (!FirstBundle)
Out << ", ";
FirstBundle = false;
Out << '"';
PrintEscapedString(BU.getTagName(), Out);
Out << '"';
Out << '(';
bool FirstInput = true;
for (const auto &Input : BU.Inputs) {
if (!FirstInput)
Out << ", ";
FirstInput = false;
TypePrinter.print(Input->getType(), Out);
Out << " ";
WriteAsOperandInternal(Out, Input, &TypePrinter, &Machine, TheModule);
}
Out << ')';
}
Out << " ]";
}
void AssemblyWriter::printModule(const Module *M) {
Machine.initialize();
if (ShouldPreserveUseListOrder)
UseListOrders = predictUseListOrder(M);
if (!M->getModuleIdentifier().empty() &&
// Don't print the ID if it will start a new line (which would
// require a comment char before it).
M->getModuleIdentifier().find('\n') == std::string::npos)
Out << "; ModuleID = '" << M->getModuleIdentifier() << "'\n";
if (!M->getSourceFileName().empty()) {
Out << "source_filename = \"";
PrintEscapedString(M->getSourceFileName(), Out);
Out << "\"\n";
}
const std::string &DL = M->getDataLayoutStr();
if (!DL.empty())
Out << "target datalayout = \"" << DL << "\"\n";
if (!M->getTargetTriple().empty())
Out << "target triple = \"" << M->getTargetTriple() << "\"\n";
if (!M->getModuleInlineAsm().empty()) {
Out << '\n';
// Split the string into lines, to make it easier to read the .ll file.
StringRef Asm = M->getModuleInlineAsm();
do {
StringRef Front;
std::tie(Front, Asm) = Asm.split('\n');
// We found a newline, print the portion of the asm string from the
// last newline up to this newline.
Out << "module asm \"";
PrintEscapedString(Front, Out);
Out << "\"\n";
} while (!Asm.empty());
}
printTypeIdentities();
// Output all comdats.
if (!Comdats.empty())
Out << '\n';
for (const Comdat *C : Comdats) {
printComdat(C);
if (C != Comdats.back())
Out << '\n';
}
// Output all globals.
if (!M->global_empty()) Out << '\n';
for (const GlobalVariable &GV : M->globals()) {
printGlobal(&GV); Out << '\n';
}
// Output all aliases.
if (!M->alias_empty()) Out << "\n";
for (const GlobalAlias &GA : M->aliases())
printIndirectSymbol(&GA);
// Output all ifuncs.
if (!M->ifunc_empty()) Out << "\n";
for (const GlobalIFunc &GI : M->ifuncs())
printIndirectSymbol(&GI);
// Output global use-lists.
printUseLists(nullptr);
// Output all of the functions.
for (const Function &F : *M)
printFunction(&F);
assert(UseListOrders.empty() && "All use-lists should have been consumed");
// Output all attribute groups.
if (!Machine.as_empty()) {
Out << '\n';
writeAllAttributeGroups();
}
2009-07-30 00:04:47 +02:00
// Output named metadata.
if (!M->named_metadata_empty()) Out << '\n';
2011-09-30 21:48:58 +02:00
for (const NamedMDNode &Node : M->named_metadata())
printNamedMDNode(&Node);
2009-07-30 00:04:47 +02:00
// Output metadata.
if (!Machine.mdn_empty()) {
Out << '\n';
writeAllMDNodes();
}
2001-06-06 22:29:01 +02:00
}
static void printMetadataIdentifier(StringRef Name,
formatted_raw_ostream &Out) {
if (Name.empty()) {
Out << "<empty name> ";
} else {
if (isalpha(static_cast<unsigned char>(Name[0])) || Name[0] == '-' ||
Name[0] == '$' || Name[0] == '.' || Name[0] == '_')
Out << Name[0];
else
Out << '\\' << hexdigit(Name[0] >> 4) << hexdigit(Name[0] & 0x0F);
for (unsigned i = 1, e = Name.size(); i != e; ++i) {
unsigned char C = Name[i];
if (isalnum(static_cast<unsigned char>(C)) || C == '-' || C == '$' ||
C == '.' || C == '_')
Out << C;
else
Out << '\\' << hexdigit(C >> 4) << hexdigit(C & 0x0F);
}
}
}
void AssemblyWriter::printNamedMDNode(const NamedMDNode *NMD) {
Out << '!';
printMetadataIdentifier(NMD->getName(), Out);
Out << " = !{";
for (unsigned i = 0, e = NMD->getNumOperands(); i != e; ++i) {
if (i)
Out << ", ";
// Write DIExpressions inline.
// FIXME: Ban DIExpressions in NamedMDNodes, they will serve no purpose.
MDNode *Op = NMD->getOperand(i);
if (auto *Expr = dyn_cast<DIExpression>(Op)) {
writeDIExpression(Out, Expr, nullptr, nullptr, nullptr);
continue;
}
int Slot = Machine.getMetadataSlot(Op);
if (Slot == -1)
Out << "<badref>";
else
Out << '!' << Slot;
}
Out << "}\n";
}
static const char *getLinkagePrintName(GlobalValue::LinkageTypes LT) {
switch (LT) {
case GlobalValue::ExternalLinkage:
return "";
case GlobalValue::PrivateLinkage:
return "private ";
case GlobalValue::InternalLinkage:
return "internal ";
case GlobalValue::LinkOnceAnyLinkage:
return "linkonce ";
case GlobalValue::LinkOnceODRLinkage:
return "linkonce_odr ";
case GlobalValue::WeakAnyLinkage:
return "weak ";
case GlobalValue::WeakODRLinkage:
return "weak_odr ";
case GlobalValue::CommonLinkage:
return "common ";
case GlobalValue::AppendingLinkage:
return "appending ";
case GlobalValue::ExternalWeakLinkage:
return "extern_weak ";
case GlobalValue::AvailableExternallyLinkage:
return "available_externally ";
}
llvm_unreachable("invalid linkage");
}
static void PrintVisibility(GlobalValue::VisibilityTypes Vis,
formatted_raw_ostream &Out) {
switch (Vis) {
case GlobalValue::DefaultVisibility: break;
case GlobalValue::HiddenVisibility: Out << "hidden "; break;
case GlobalValue::ProtectedVisibility: Out << "protected "; break;
}
}
static void PrintDSOLocation(const GlobalValue &GV,
formatted_raw_ostream &Out) {
// GVs with local linkage or non default visibility are implicitly dso_local,
// so we don't print it.
bool Implicit = GV.hasLocalLinkage() ||
(!GV.hasExternalWeakLinkage() && !GV.hasDefaultVisibility());
if (GV.isDSOLocal() && !Implicit)
Represent runtime preemption in the IR. Currently we do not represent runtime preemption in the IR, which has several drawbacks: 1) The semantics of GlobalValues differ depending on the object file format you are targeting (as well as the relocation-model and -fPIE value). 2) We have no way of disabling inlining of run time interposable functions, since in the IR we only know if a function is link-time interposable. Because of this llvm cannot support elf-interposition semantics. 3) In LTO builds of executables we will have extra knowledge that a symbol resolved to a local definition and can't be preemptable, but have no way to propagate that knowledge through the compiler. This patch adds preemptability specifiers to the IR with the following meaning: dso_local --> means the compiler may assume the symbol will resolve to a definition within the current linkage unit and the symbol may be accessed directly even if the definition is not within this compilation unit. dso_preemptable --> means that the compiler must assume the GlobalValue may be replaced with a definition from outside the current linkage unit at runtime. To ease transitioning dso_preemptable is treated as a 'default' in that low-level codegen will still do the same checks it did previously to see if a symbol should be accessed indirectly. Eventually when IR producers emit the specifiers on all Globalvalues we can change dso_preemptable to mean 'always access indirectly', and remove the current logic. Differential Revision: https://reviews.llvm.org/D20217 llvm-svn: 316668
2017-10-26 17:00:26 +02:00
Out << "dso_local ";
}
static void PrintDLLStorageClass(GlobalValue::DLLStorageClassTypes SCT,
formatted_raw_ostream &Out) {
switch (SCT) {
case GlobalValue::DefaultStorageClass: break;
case GlobalValue::DLLImportStorageClass: Out << "dllimport "; break;
case GlobalValue::DLLExportStorageClass: Out << "dllexport "; break;
}
}
static void PrintThreadLocalModel(GlobalVariable::ThreadLocalMode TLM,
formatted_raw_ostream &Out) {
switch (TLM) {
case GlobalVariable::NotThreadLocal:
break;
case GlobalVariable::GeneralDynamicTLSModel:
Out << "thread_local ";
break;
case GlobalVariable::LocalDynamicTLSModel:
Out << "thread_local(localdynamic) ";
break;
case GlobalVariable::InitialExecTLSModel:
Out << "thread_local(initialexec) ";
break;
case GlobalVariable::LocalExecTLSModel:
Out << "thread_local(localexec) ";
break;
}
}
IR: Introduce local_unnamed_addr attribute. If a local_unnamed_addr attribute is attached to a global, the address is known to be insignificant within the module. It is distinct from the existing unnamed_addr attribute in that it only describes a local property of the module rather than a global property of the symbol. This attribute is intended to be used by the code generator and LTO to allow the linker to decide whether the global needs to be in the symbol table. It is possible to exclude a global from the symbol table if three things are true: - This attribute is present on every instance of the global (which means that the normal rule that the global must have a unique address can be broken without being observable by the program by performing comparisons against the global's address) - The global has linkonce_odr linkage (which means that each linkage unit must have its own copy of the global if it requires one, and the copy in each linkage unit must be the same) - It is a constant or a function (which means that the program cannot observe that the unique-address rule has been broken by writing to the global) Although this attribute could in principle be computed from the module contents, LTO clients (i.e. linkers) will normally need to be able to compute this property as part of symbol resolution, and it would be inefficient to materialize every module just to compute it. See: http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20160509/356401.html http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20160516/356738.html for earlier discussion. Part of the fix for PR27553. Differential Revision: http://reviews.llvm.org/D20348 llvm-svn: 272709
2016-06-14 23:01:22 +02:00
static StringRef getUnnamedAddrEncoding(GlobalVariable::UnnamedAddr UA) {
switch (UA) {
case GlobalVariable::UnnamedAddr::None:
return "";
case GlobalVariable::UnnamedAddr::Local:
return "local_unnamed_addr";
case GlobalVariable::UnnamedAddr::Global:
return "unnamed_addr";
}
llvm_unreachable("Unknown UnnamedAddr");
IR: Introduce local_unnamed_addr attribute. If a local_unnamed_addr attribute is attached to a global, the address is known to be insignificant within the module. It is distinct from the existing unnamed_addr attribute in that it only describes a local property of the module rather than a global property of the symbol. This attribute is intended to be used by the code generator and LTO to allow the linker to decide whether the global needs to be in the symbol table. It is possible to exclude a global from the symbol table if three things are true: - This attribute is present on every instance of the global (which means that the normal rule that the global must have a unique address can be broken without being observable by the program by performing comparisons against the global's address) - The global has linkonce_odr linkage (which means that each linkage unit must have its own copy of the global if it requires one, and the copy in each linkage unit must be the same) - It is a constant or a function (which means that the program cannot observe that the unique-address rule has been broken by writing to the global) Although this attribute could in principle be computed from the module contents, LTO clients (i.e. linkers) will normally need to be able to compute this property as part of symbol resolution, and it would be inefficient to materialize every module just to compute it. See: http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20160509/356401.html http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20160516/356738.html for earlier discussion. Part of the fix for PR27553. Differential Revision: http://reviews.llvm.org/D20348 llvm-svn: 272709
2016-06-14 23:01:22 +02:00
}
static void maybePrintComdat(formatted_raw_ostream &Out,
const GlobalObject &GO) {
const Comdat *C = GO.getComdat();
if (!C)
return;
if (isa<GlobalVariable>(GO))
Out << ',';
Out << " comdat";
if (GO.getName() == C->getName())
return;
Out << '(';
PrintLLVMName(Out, C->getName(), ComdatPrefix);
Out << ')';
}
void AssemblyWriter::printGlobal(const GlobalVariable *GV) {
if (GV->isMaterializable())
Out << "; Materializable\n";
WriteAsOperandInternal(Out, GV, &TypePrinter, &Machine, GV->getParent());
Out << " = ";
if (!GV->hasInitializer() && GV->hasExternalLinkage())
Out << "external ";
Out << getLinkagePrintName(GV->getLinkage());
PrintDSOLocation(*GV, Out);
PrintVisibility(GV->getVisibility(), Out);
PrintDLLStorageClass(GV->getDLLStorageClass(), Out);
PrintThreadLocalModel(GV->getThreadLocalMode(), Out);
IR: Introduce local_unnamed_addr attribute. If a local_unnamed_addr attribute is attached to a global, the address is known to be insignificant within the module. It is distinct from the existing unnamed_addr attribute in that it only describes a local property of the module rather than a global property of the symbol. This attribute is intended to be used by the code generator and LTO to allow the linker to decide whether the global needs to be in the symbol table. It is possible to exclude a global from the symbol table if three things are true: - This attribute is present on every instance of the global (which means that the normal rule that the global must have a unique address can be broken without being observable by the program by performing comparisons against the global's address) - The global has linkonce_odr linkage (which means that each linkage unit must have its own copy of the global if it requires one, and the copy in each linkage unit must be the same) - It is a constant or a function (which means that the program cannot observe that the unique-address rule has been broken by writing to the global) Although this attribute could in principle be computed from the module contents, LTO clients (i.e. linkers) will normally need to be able to compute this property as part of symbol resolution, and it would be inefficient to materialize every module just to compute it. See: http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20160509/356401.html http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20160516/356738.html for earlier discussion. Part of the fix for PR27553. Differential Revision: http://reviews.llvm.org/D20348 llvm-svn: 272709
2016-06-14 23:01:22 +02:00
StringRef UA = getUnnamedAddrEncoding(GV->getUnnamedAddr());
if (!UA.empty())
Out << UA << ' ';
if (unsigned AddressSpace = GV->getType()->getAddressSpace())
Out << "addrspace(" << AddressSpace << ") ";
if (GV->isExternallyInitialized()) Out << "externally_initialized ";
Out << (GV->isConstant() ? "constant " : "global ");
TypePrinter.print(GV->getValueType(), Out);
if (GV->hasInitializer()) {
Out << ' ';
writeOperand(GV->getInitializer(), false);
}
if (GV->hasSection()) {
Out << ", section \"";
PrintEscapedString(GV->getSection(), Out);
Out << '"';
}
maybePrintComdat(Out, *GV);
2005-11-12 01:10:19 +01:00
if (GV->getAlignment())
Out << ", align " << GV->getAlignment();
SmallVector<std::pair<unsigned, MDNode *>, 4> MDs;
GV->getAllMetadata(MDs);
printMetadataAttachments(MDs, ", ");
auto Attrs = GV->getAttributes();
if (Attrs.hasAttributes())
Out << " #" << Machine.getAttributeGroupSlot(Attrs);
printInfoComment(*GV);
}
void AssemblyWriter::printIndirectSymbol(const GlobalIndirectSymbol *GIS) {
if (GIS->isMaterializable())
Out << "; Materializable\n";
WriteAsOperandInternal(Out, GIS, &TypePrinter, &Machine, GIS->getParent());
Out << " = ";
Out << getLinkagePrintName(GIS->getLinkage());
PrintDSOLocation(*GIS, Out);
PrintVisibility(GIS->getVisibility(), Out);
PrintDLLStorageClass(GIS->getDLLStorageClass(), Out);
PrintThreadLocalModel(GIS->getThreadLocalMode(), Out);
IR: Introduce local_unnamed_addr attribute. If a local_unnamed_addr attribute is attached to a global, the address is known to be insignificant within the module. It is distinct from the existing unnamed_addr attribute in that it only describes a local property of the module rather than a global property of the symbol. This attribute is intended to be used by the code generator and LTO to allow the linker to decide whether the global needs to be in the symbol table. It is possible to exclude a global from the symbol table if three things are true: - This attribute is present on every instance of the global (which means that the normal rule that the global must have a unique address can be broken without being observable by the program by performing comparisons against the global's address) - The global has linkonce_odr linkage (which means that each linkage unit must have its own copy of the global if it requires one, and the copy in each linkage unit must be the same) - It is a constant or a function (which means that the program cannot observe that the unique-address rule has been broken by writing to the global) Although this attribute could in principle be computed from the module contents, LTO clients (i.e. linkers) will normally need to be able to compute this property as part of symbol resolution, and it would be inefficient to materialize every module just to compute it. See: http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20160509/356401.html http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20160516/356738.html for earlier discussion. Part of the fix for PR27553. Differential Revision: http://reviews.llvm.org/D20348 llvm-svn: 272709
2016-06-14 23:01:22 +02:00
StringRef UA = getUnnamedAddrEncoding(GIS->getUnnamedAddr());
if (!UA.empty())
Out << UA << ' ';
if (isa<GlobalAlias>(GIS))
Out << "alias ";
else if (isa<GlobalIFunc>(GIS))
Out << "ifunc ";
else
llvm_unreachable("Not an alias or ifunc!");
TypePrinter.print(GIS->getValueType(), Out);
[opaque pointer type] Add textual IR support for explicit type parameter for global aliases update.py: import fileinput import sys import re alias_match_prefix = r"(.*(?:=|:|^)\s*(?:external |)(?:(?:private|internal|linkonce|linkonce_odr|weak|weak_odr|common|appending|extern_weak|available_externally) )?(?:default |hidden |protected )?(?:dllimport |dllexport )?(?:unnamed_addr |)(?:thread_local(?:\([a-z]*\))? )?alias" plain = re.compile(alias_match_prefix + r" (.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|addrspacecast|\[\[[a-zA-Z]|\{\{).*$)") cast = re.compile(alias_match_prefix + r") ((?:bitcast|inttoptr|addrspacecast)\s*\(.* to (.*?)(| addrspace\(\d+\) *)\*\)\s*(?:;.*)?$)") gep = re.compile(alias_match_prefix + r") ((?:getelementptr)\s*(?:inbounds)?\s*\((?P<type>.*), (?P=type)(?:\s*addrspace\(\d+\)\s*)?\* .*\)\s*(?:;.*)?$)") def conv(line): m = re.match(cast, line) if m: return m.group(1) + " " + m.group(3) + ", " + m.group(2) m = re.match(gep, line) if m: return m.group(1) + " " + m.group(3) + ", " + m.group(2) m = re.match(plain, line) if m: return m.group(1) + ", " + m.group(2) + m.group(3) + "*" + m.group(4) + "\n" return line for line in sys.stdin: sys.stdout.write(conv(line)) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh llvm-svn: 247378
2015-09-11 05:22:04 +02:00
Out << ", ";
const Constant *IS = GIS->getIndirectSymbol();
if (!IS) {
TypePrinter.print(GIS->getType(), Out);
Out << " <<NULL ALIASEE>>";
2011-08-01 14:48:54 +02:00
} else {
writeOperand(IS, !isa<ConstantExpr>(IS));
2011-08-01 14:48:54 +02:00
}
printInfoComment(*GIS);
Out << '\n';
}
void AssemblyWriter::printComdat(const Comdat *C) {
C->print(Out);
}
void AssemblyWriter::printTypeIdentities() {
[MIR] Making MIR Printing, opt -dot-cfg, and -debug printing faster Value::printAsOperand has been scanning the entire module just to print a single value as an operand, regardless being asked to print a type or not at all, and regardless really needing to scan the module to print a type. It made some of the users of the method exceptionally slow on large IR-modules (or large MIR-files with large IR-modules embedded). This patch defers scanning a module looking for struct types, mostly numbered struct types, as much as possible, speeding up those users w/o changing any APIs at all. See speedup examples below: Release Build: # 83 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 133 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output Release + Asserts Build: # 95 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 146 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output # 1096 seconds -> 553 seconds time ./bin/llc -debug-only=isel -fast-isel=false -stop-after=isel \ sqlite3.O0.ll -o /dev/null 2> err where sqlite3.O0.ll is non-optimized IR produced from sqlite-amalgamation (http://sqlite.org/download.html), which is entire SQLite3 implementation in a single C-file. Benchmarked on 4-cores / 8 threads PCI-E SSD iMac running macOS Reviewers: dexonsmith, bkramer, void, chandlerc, aditya_nandakumar, dsanders, qcolombet, Reviewed By: bogner Subscribers: thegameg, llvm-commits Differential Revision: https://reviews.llvm.org/D44132 llvm-svn: 328246
2018-03-22 22:29:07 +01:00
if (TypePrinter.empty())
return;
2011-09-30 21:48:58 +02:00
Out << '\n';
2011-09-30 21:48:58 +02:00
// Emit all numbered types.
[MIR] Making MIR Printing, opt -dot-cfg, and -debug printing faster Value::printAsOperand has been scanning the entire module just to print a single value as an operand, regardless being asked to print a type or not at all, and regardless really needing to scan the module to print a type. It made some of the users of the method exceptionally slow on large IR-modules (or large MIR-files with large IR-modules embedded). This patch defers scanning a module looking for struct types, mostly numbered struct types, as much as possible, speeding up those users w/o changing any APIs at all. See speedup examples below: Release Build: # 83 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 133 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output Release + Asserts Build: # 95 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 146 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output # 1096 seconds -> 553 seconds time ./bin/llc -debug-only=isel -fast-isel=false -stop-after=isel \ sqlite3.O0.ll -o /dev/null 2> err where sqlite3.O0.ll is non-optimized IR produced from sqlite-amalgamation (http://sqlite.org/download.html), which is entire SQLite3 implementation in a single C-file. Benchmarked on 4-cores / 8 threads PCI-E SSD iMac running macOS Reviewers: dexonsmith, bkramer, void, chandlerc, aditya_nandakumar, dsanders, qcolombet, Reviewed By: bogner Subscribers: thegameg, llvm-commits Differential Revision: https://reviews.llvm.org/D44132 llvm-svn: 328246
2018-03-22 22:29:07 +01:00
auto &NumberedTypes = TypePrinter.getNumberedTypes();
for (unsigned I = 0, E = NumberedTypes.size(); I != E; ++I) {
Out << '%' << I << " = type ";
2011-09-30 21:48:58 +02:00
// Make sure we print out at least one level of the type structure, so
// that we do not get %2 = type %2
[MIR] Making MIR Printing, opt -dot-cfg, and -debug printing faster Value::printAsOperand has been scanning the entire module just to print a single value as an operand, regardless being asked to print a type or not at all, and regardless really needing to scan the module to print a type. It made some of the users of the method exceptionally slow on large IR-modules (or large MIR-files with large IR-modules embedded). This patch defers scanning a module looking for struct types, mostly numbered struct types, as much as possible, speeding up those users w/o changing any APIs at all. See speedup examples below: Release Build: # 83 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 133 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output Release + Asserts Build: # 95 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 146 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output # 1096 seconds -> 553 seconds time ./bin/llc -debug-only=isel -fast-isel=false -stop-after=isel \ sqlite3.O0.ll -o /dev/null 2> err where sqlite3.O0.ll is non-optimized IR produced from sqlite-amalgamation (http://sqlite.org/download.html), which is entire SQLite3 implementation in a single C-file. Benchmarked on 4-cores / 8 threads PCI-E SSD iMac running macOS Reviewers: dexonsmith, bkramer, void, chandlerc, aditya_nandakumar, dsanders, qcolombet, Reviewed By: bogner Subscribers: thegameg, llvm-commits Differential Revision: https://reviews.llvm.org/D44132 llvm-svn: 328246
2018-03-22 22:29:07 +01:00
TypePrinter.printStructBody(NumberedTypes[I], Out);
Out << '\n';
}
2011-09-30 21:48:58 +02:00
[MIR] Making MIR Printing, opt -dot-cfg, and -debug printing faster Value::printAsOperand has been scanning the entire module just to print a single value as an operand, regardless being asked to print a type or not at all, and regardless really needing to scan the module to print a type. It made some of the users of the method exceptionally slow on large IR-modules (or large MIR-files with large IR-modules embedded). This patch defers scanning a module looking for struct types, mostly numbered struct types, as much as possible, speeding up those users w/o changing any APIs at all. See speedup examples below: Release Build: # 83 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 133 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output Release + Asserts Build: # 95 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 146 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output # 1096 seconds -> 553 seconds time ./bin/llc -debug-only=isel -fast-isel=false -stop-after=isel \ sqlite3.O0.ll -o /dev/null 2> err where sqlite3.O0.ll is non-optimized IR produced from sqlite-amalgamation (http://sqlite.org/download.html), which is entire SQLite3 implementation in a single C-file. Benchmarked on 4-cores / 8 threads PCI-E SSD iMac running macOS Reviewers: dexonsmith, bkramer, void, chandlerc, aditya_nandakumar, dsanders, qcolombet, Reviewed By: bogner Subscribers: thegameg, llvm-commits Differential Revision: https://reviews.llvm.org/D44132 llvm-svn: 328246
2018-03-22 22:29:07 +01:00
auto &NamedTypes = TypePrinter.getNamedTypes();
for (unsigned I = 0, E = NamedTypes.size(); I != E; ++I) {
PrintLLVMName(Out, NamedTypes[I]->getName(), LocalPrefix);
Out << " = type ";
// Make sure we print out at least one level of the type structure, so
// that we do not get %FILE = type %FILE
[MIR] Making MIR Printing, opt -dot-cfg, and -debug printing faster Value::printAsOperand has been scanning the entire module just to print a single value as an operand, regardless being asked to print a type or not at all, and regardless really needing to scan the module to print a type. It made some of the users of the method exceptionally slow on large IR-modules (or large MIR-files with large IR-modules embedded). This patch defers scanning a module looking for struct types, mostly numbered struct types, as much as possible, speeding up those users w/o changing any APIs at all. See speedup examples below: Release Build: # 83 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 133 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output Release + Asserts Build: # 95 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 146 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output # 1096 seconds -> 553 seconds time ./bin/llc -debug-only=isel -fast-isel=false -stop-after=isel \ sqlite3.O0.ll -o /dev/null 2> err where sqlite3.O0.ll is non-optimized IR produced from sqlite-amalgamation (http://sqlite.org/download.html), which is entire SQLite3 implementation in a single C-file. Benchmarked on 4-cores / 8 threads PCI-E SSD iMac running macOS Reviewers: dexonsmith, bkramer, void, chandlerc, aditya_nandakumar, dsanders, qcolombet, Reviewed By: bogner Subscribers: thegameg, llvm-commits Differential Revision: https://reviews.llvm.org/D44132 llvm-svn: 328246
2018-03-22 22:29:07 +01:00
TypePrinter.printStructBody(NamedTypes[I], Out);
Out << '\n';
}
}
2004-03-02 01:22:19 +01:00
/// printFunction - Print all aspects of a function.
void AssemblyWriter::printFunction(const Function *F) {
// Print out the return type and name.
Out << '\n';
if (AnnotationWriter) AnnotationWriter->emitFunctionAnnot(F, Out);
if (F->isMaterializable())
Out << "; Materializable\n";
const AttributeList &Attrs = F->getAttributes();
if (Attrs.hasAttributes(AttributeList::FunctionIndex)) {
AttributeSet AS = Attrs.getFnAttributes();
std::string AttrStr;
for (const Attribute &Attr : AS) {
if (!Attr.isStringAttribute()) {
if (!AttrStr.empty()) AttrStr += ' ';
AttrStr += Attr.getAsString();
}
}
if (!AttrStr.empty())
Out << "; Function Attrs: " << AttrStr << '\n';
}
IR: Allow metadata attachments on declarations, and fix lazy loaded metadata issue with globals. This change is motivated by an upcoming change to the metadata representation used for CFI. The indirect function call checker needs type information for external function declarations in order to correctly generate jump table entries for such declarations. We currently associate such type information with declarations using a global metadata node, but I plan [1] to move all such metadata to global object attachments. In bitcode, metadata attachments for function declarations appear in the global metadata block. This seems reasonable to me because I expect metadata attachments on declarations to be uncommon. In the long term I'd also expect this to be the case for CFI, because we'd want to use some specialized bitcode format for this metadata that could be read as part of the ThinLTO thin-link phase, which would mean that it would not appear in the global metadata block. To solve the lazy loaded metadata issue I was seeing with D20147, I use the same bitcode representation for metadata attachments for global variables as I do for function declarations. Since there's a use case for metadata attachments in the global metadata block, we might as well use that representation for global variables as well, at least until we have a mechanism for lazy loading global variables. In the assembly format, the metadata attachments appear after the "declare" keyword in order to avoid a parsing ambiguity. [1] http://lists.llvm.org/pipermail/llvm-dev/2016-June/100462.html Differential Revision: http://reviews.llvm.org/D21052 llvm-svn: 273336
2016-06-22 01:42:48 +02:00
Machine.incorporateFunction(F);
if (F->isDeclaration()) {
Out << "declare";
SmallVector<std::pair<unsigned, MDNode *>, 4> MDs;
F->getAllMetadata(MDs);
printMetadataAttachments(MDs, " ");
Out << ' ';
} else
Out << "define ";
Out << getLinkagePrintName(F->getLinkage());
PrintDSOLocation(*F, Out);
PrintVisibility(F->getVisibility(), Out);
PrintDLLStorageClass(F->getDLLStorageClass(), Out);
// Print the calling convention.
if (F->getCallingConv() != CallingConv::C) {
PrintCallingConv(F->getCallingConv(), Out);
Out << " ";
}
FunctionType *FT = F->getFunctionType();
if (Attrs.hasAttributes(AttributeList::ReturnIndex))
Out << Attrs.getAsString(AttributeList::ReturnIndex) << ' ';
TypePrinter.print(F->getReturnType(), Out);
Out << ' ';
WriteAsOperandInternal(Out, F, &TypePrinter, &Machine, F->getParent());
Out << '(';
// Loop over the arguments, printing them...
if (F->isDeclaration() && !IsForDebug) {
// We're only interested in the type here - don't print argument names.
for (unsigned I = 0, E = FT->getNumParams(); I != E; ++I) {
// Insert commas as we go... the first arg doesn't get a comma
if (I)
Out << ", ";
// Output type...
TypePrinter.print(FT->getParamType(I), Out);
AttributeSet ArgAttrs = Attrs.getParamAttributes(I);
if (ArgAttrs.hasAttributes())
Out << ' ' << ArgAttrs.getAsString();
}
} else {
// The arguments are meaningful here, print them in detail.
for (const Argument &Arg : F->args()) {
// Insert commas as we go... the first arg doesn't get a comma
if (Arg.getArgNo() != 0)
Out << ", ";
printArgument(&Arg, Attrs.getParamAttributes(Arg.getArgNo()));
}
}
// Finish printing arguments...
if (FT->isVarArg()) {
if (FT->getNumParams()) Out << ", ";
Out << "..."; // Output varargs portion of signature!
}
Out << ')';
IR: Introduce local_unnamed_addr attribute. If a local_unnamed_addr attribute is attached to a global, the address is known to be insignificant within the module. It is distinct from the existing unnamed_addr attribute in that it only describes a local property of the module rather than a global property of the symbol. This attribute is intended to be used by the code generator and LTO to allow the linker to decide whether the global needs to be in the symbol table. It is possible to exclude a global from the symbol table if three things are true: - This attribute is present on every instance of the global (which means that the normal rule that the global must have a unique address can be broken without being observable by the program by performing comparisons against the global's address) - The global has linkonce_odr linkage (which means that each linkage unit must have its own copy of the global if it requires one, and the copy in each linkage unit must be the same) - It is a constant or a function (which means that the program cannot observe that the unique-address rule has been broken by writing to the global) Although this attribute could in principle be computed from the module contents, LTO clients (i.e. linkers) will normally need to be able to compute this property as part of symbol resolution, and it would be inefficient to materialize every module just to compute it. See: http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20160509/356401.html http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20160516/356738.html for earlier discussion. Part of the fix for PR27553. Differential Revision: http://reviews.llvm.org/D20348 llvm-svn: 272709
2016-06-14 23:01:22 +02:00
StringRef UA = getUnnamedAddrEncoding(F->getUnnamedAddr());
if (!UA.empty())
Out << ' ' << UA;
if (Attrs.hasAttributes(AttributeList::FunctionIndex))
Out << " #" << Machine.getAttributeGroupSlot(Attrs.getFnAttributes());
if (F->hasSection()) {
Out << " section \"";
PrintEscapedString(F->getSection(), Out);
Out << '"';
}
maybePrintComdat(Out, *F);
if (F->getAlignment())
Out << " align " << F->getAlignment();
if (F->hasGC())
Out << " gc \"" << F->getGC() << '"';
if (F->hasPrefixData()) {
Out << " prefix ";
writeOperand(F->getPrefixData(), true);
}
Prologue support Patch by Ben Gamari! This redefines the `prefix` attribute introduced previously and introduces a `prologue` attribute. There are a two primary usecases that these attributes aim to serve, 1. Function prologue sigils 2. Function hot-patching: Enable the user to insert `nop` operations at the beginning of the function which can later be safely replaced with a call to some instrumentation facility 3. Runtime metadata: Allow a compiler to insert data for use by the runtime during execution. GHC is one example of a compiler that needs this functionality for its tables-next-to-code functionality. Previously `prefix` served cases (1) and (2) quite well by allowing the user to introduce arbitrary data at the entrypoint but before the function body. Case (3), however, was poorly handled by this approach as it required that prefix data was valid executable code. Here we redefine the notion of prefix data to instead be data which occurs immediately before the function entrypoint (i.e. the symbol address). Since prefix data now occurs before the function entrypoint, there is no need for the data to be valid code. The previous notion of prefix data now goes under the name "prologue data" to emphasize its duality with the function epilogue. The intention here is to handle cases (1) and (2) with prologue data and case (3) with prefix data. References ---------- This idea arose out of discussions[1] with Reid Kleckner in response to a proposal to introduce the notion of symbol offsets to enable handling of case (3). [1] http://lists.cs.uiuc.edu/pipermail/llvmdev/2014-May/073235.html Test Plan: testsuite Differential Revision: http://reviews.llvm.org/D6454 llvm-svn: 223189
2014-12-03 03:08:38 +01:00
if (F->hasPrologueData()) {
Out << " prologue ";
writeOperand(F->getPrologueData(), true);
}
if (F->hasPersonalityFn()) {
Out << " personality ";
writeOperand(F->getPersonalityFn(), /*PrintType=*/true);
}
Prologue support Patch by Ben Gamari! This redefines the `prefix` attribute introduced previously and introduces a `prologue` attribute. There are a two primary usecases that these attributes aim to serve, 1. Function prologue sigils 2. Function hot-patching: Enable the user to insert `nop` operations at the beginning of the function which can later be safely replaced with a call to some instrumentation facility 3. Runtime metadata: Allow a compiler to insert data for use by the runtime during execution. GHC is one example of a compiler that needs this functionality for its tables-next-to-code functionality. Previously `prefix` served cases (1) and (2) quite well by allowing the user to introduce arbitrary data at the entrypoint but before the function body. Case (3), however, was poorly handled by this approach as it required that prefix data was valid executable code. Here we redefine the notion of prefix data to instead be data which occurs immediately before the function entrypoint (i.e. the symbol address). Since prefix data now occurs before the function entrypoint, there is no need for the data to be valid code. The previous notion of prefix data now goes under the name "prologue data" to emphasize its duality with the function epilogue. The intention here is to handle cases (1) and (2) with prologue data and case (3) with prefix data. References ---------- This idea arose out of discussions[1] with Reid Kleckner in response to a proposal to introduce the notion of symbol offsets to enable handling of case (3). [1] http://lists.cs.uiuc.edu/pipermail/llvmdev/2014-May/073235.html Test Plan: testsuite Differential Revision: http://reviews.llvm.org/D6454 llvm-svn: 223189
2014-12-03 03:08:38 +01:00
if (F->isDeclaration()) {
Out << '\n';
} else {
IR: Allow metadata attachments on declarations, and fix lazy loaded metadata issue with globals. This change is motivated by an upcoming change to the metadata representation used for CFI. The indirect function call checker needs type information for external function declarations in order to correctly generate jump table entries for such declarations. We currently associate such type information with declarations using a global metadata node, but I plan [1] to move all such metadata to global object attachments. In bitcode, metadata attachments for function declarations appear in the global metadata block. This seems reasonable to me because I expect metadata attachments on declarations to be uncommon. In the long term I'd also expect this to be the case for CFI, because we'd want to use some specialized bitcode format for this metadata that could be read as part of the ThinLTO thin-link phase, which would mean that it would not appear in the global metadata block. To solve the lazy loaded metadata issue I was seeing with D20147, I use the same bitcode representation for metadata attachments for global variables as I do for function declarations. Since there's a use case for metadata attachments in the global metadata block, we might as well use that representation for global variables as well, at least until we have a mechanism for lazy loading global variables. In the assembly format, the metadata attachments appear after the "declare" keyword in order to avoid a parsing ambiguity. [1] http://lists.llvm.org/pipermail/llvm-dev/2016-June/100462.html Differential Revision: http://reviews.llvm.org/D21052 llvm-svn: 273336
2016-06-22 01:42:48 +02:00
SmallVector<std::pair<unsigned, MDNode *>, 4> MDs;
F->getAllMetadata(MDs);
printMetadataAttachments(MDs, " ");
Out << " {";
// Output all of the function's basic blocks.
for (const BasicBlock &BB : *F)
printBasicBlock(&BB);
// Output the function's use-lists.
printUseLists(F);
Out << "}\n";
}
Machine.purgeFunction();
2001-06-06 22:29:01 +02:00
}
2004-03-02 01:22:19 +01:00
/// printArgument - This member is called for every argument that is passed into
/// the function. Simply print it out
void AssemblyWriter::printArgument(const Argument *Arg, AttributeSet Attrs) {
2001-06-06 22:29:01 +02:00
// Output type...
TypePrinter.print(Arg->getType(), Out);
// Output parameter attributes list
if (Attrs.hasAttributes())
Out << ' ' << Attrs.getAsString();
2001-06-06 22:29:01 +02:00
// Output name, if available...
if (Arg->hasName()) {
Out << ' ';
PrintLLVMName(Out, Arg);
}
2001-06-06 22:29:01 +02:00
}
2004-03-02 01:22:19 +01:00
/// printBasicBlock - This member is called for each basic block in a method.
void AssemblyWriter::printBasicBlock(const BasicBlock *BB) {
if (BB->hasName()) { // Print out the label if it exists...
Out << "\n";
PrintLLVMName(Out, BB->getName(), LabelPrefix);
Out << ':';
} else if (!BB->use_empty()) { // Don't print block # of no uses...
Out << "\n; <label>:";
int Slot = Machine.getLocalSlot(BB);
if (Slot != -1)
Out << Slot << ":";
else
Out << "<badref>";
}
if (!BB->getParent()) {
Out.PadToColumn(50);
Out << "; Error: Block without parent!";
} else if (BB != &BB->getParent()->getEntryBlock()) { // Not the entry block?
// Output predecessors for the block.
Out.PadToColumn(50);
Out << ";";
const_pred_iterator PI = pred_begin(BB), PE = pred_end(BB);
if (PI == PE) {
Out << " No predecessors!";
} else {
Out << " preds = ";
writeOperand(*PI, false);
for (++PI; PI != PE; ++PI) {
Out << ", ";
writeOperand(*PI, false);
}
}
2001-06-06 22:29:01 +02:00
}
Out << "\n";
2001-06-06 22:29:01 +02:00
if (AnnotationWriter) AnnotationWriter->emitBasicBlockStartAnnot(BB, Out);
// Output all of the instructions in the basic block...
for (const Instruction &I : *BB) {
printInstructionLine(I);
}
if (AnnotationWriter) AnnotationWriter->emitBasicBlockEndAnnot(BB, Out);
2001-06-06 22:29:01 +02:00
}
/// printInstructionLine - Print an instruction and a newline character.
void AssemblyWriter::printInstructionLine(const Instruction &I) {
printInstruction(I);
Out << '\n';
}
/// printGCRelocateComment - print comment after call to the gc.relocate
/// intrinsic indicating base and derived pointer names.
void AssemblyWriter::printGCRelocateComment(const GCRelocateInst &Relocate) {
Out << " ; (";
writeOperand(Relocate.getBasePtr(), false);
Out << ", ";
writeOperand(Relocate.getDerivedPtr(), false);
Out << ")";
}
2004-03-02 01:22:19 +01:00
/// printInfoComment - Print a little comment after the instruction indicating
/// which slot it occupies.
void AssemblyWriter::printInfoComment(const Value &V) {
if (const auto *Relocate = dyn_cast<GCRelocateInst>(&V))
printGCRelocateComment(*Relocate);
if (AnnotationWriter)
AnnotationWriter->printInfoComment(V, Out);
}
// This member is called for each Instruction in a function..
void AssemblyWriter::printInstruction(const Instruction &I) {
if (AnnotationWriter) AnnotationWriter->emitInstructionAnnot(&I, Out);
// Print out indentation for an instruction.
Out << " ";
2001-06-06 22:29:01 +02:00
// Print out name if it exists...
if (I.hasName()) {
PrintLLVMName(Out, &I);
Out << " = ";
} else if (!I.getType()->isVoidTy()) {
// Print out the def slot taken.
int SlotNum = Machine.getLocalSlot(&I);
if (SlotNum == -1)
Out << "<badref> = ";
else
Out << '%' << SlotNum << " = ";
}
2011-09-30 21:48:58 +02:00
if (const CallInst *CI = dyn_cast<CallInst>(&I)) {
if (CI->isMustTailCall())
Out << "musttail ";
else if (CI->isTailCall())
Out << "tail ";
else if (CI->isNoTailCall())
Out << "notail ";
}
2001-06-06 22:29:01 +02:00
// Print out the opcode...
Out << I.getOpcodeName();
2001-06-06 22:29:01 +02:00
// If this is an atomic load or store, print out the atomic marker.
if ((isa<LoadInst>(I) && cast<LoadInst>(I).isAtomic()) ||
(isa<StoreInst>(I) && cast<StoreInst>(I).isAtomic()))
Out << " atomic";
if (isa<AtomicCmpXchgInst>(I) && cast<AtomicCmpXchgInst>(I).isWeak())
Out << " weak";
// If this is a volatile operation, print out the volatile marker.
if ((isa<LoadInst>(I) && cast<LoadInst>(I).isVolatile()) ||
(isa<StoreInst>(I) && cast<StoreInst>(I).isVolatile()) ||
(isa<AtomicCmpXchgInst>(I) && cast<AtomicCmpXchgInst>(I).isVolatile()) ||
(isa<AtomicRMWInst>(I) && cast<AtomicRMWInst>(I).isVolatile()))
Out << " volatile";
// Print out optimization information.
WriteOptimizationInfo(Out, &I);
// Print out the compare instruction predicates
if (const CmpInst *CI = dyn_cast<CmpInst>(&I))
Out << ' ' << CmpInst::getPredicateName(CI->getPredicate());
// Print out the atomicrmw operation
if (const AtomicRMWInst *RMWI = dyn_cast<AtomicRMWInst>(&I))
writeAtomicRMWOperation(Out, RMWI->getOperation());
2001-06-06 22:29:01 +02:00
// Print out the type of the operands...
const Value *Operand = I.getNumOperands() ? I.getOperand(0) : nullptr;
2001-06-06 22:29:01 +02:00
// Special case conditional branches to swizzle the condition out to the front
if (isa<BranchInst>(I) && cast<BranchInst>(I).isConditional()) {
const BranchInst &BI(cast<BranchInst>(I));
Out << ' ';
writeOperand(BI.getCondition(), true);
Out << ", ";
writeOperand(BI.getSuccessor(0), true);
Out << ", ";
writeOperand(BI.getSuccessor(1), true);
2001-06-06 22:29:01 +02:00
} else if (isa<SwitchInst>(I)) {
const SwitchInst& SI(cast<SwitchInst>(I));
// Special case switch instruction to get formatting nice and correct.
Out << ' ';
writeOperand(SI.getCondition(), true);
Out << ", ";
writeOperand(SI.getDefaultDest(), true);
2008-08-24 00:52:27 +02:00
Out << " [";
for (auto Case : SI.cases()) {
Out << "\n ";
writeOperand(Case.getCaseValue(), true);
Out << ", ";
writeOperand(Case.getCaseSuccessor(), true);
2001-06-06 22:29:01 +02:00
}
Out << "\n ]";
} else if (isa<IndirectBrInst>(I)) {
// Special case indirectbr instruction to get formatting nice and correct.
Out << ' ';
writeOperand(Operand, true);
Out << ", [";
2011-09-30 21:48:58 +02:00
for (unsigned i = 1, e = I.getNumOperands(); i != e; ++i) {
if (i != 1)
Out << ", ";
writeOperand(I.getOperand(i), true);
}
Out << ']';
} else if (const PHINode *PN = dyn_cast<PHINode>(&I)) {
Out << ' ';
TypePrinter.print(I.getType(), Out);
Out << ' ';
2001-06-06 22:29:01 +02:00
for (unsigned op = 0, Eop = PN->getNumIncomingValues(); op < Eop; ++op) {
if (op) Out << ", ";
Out << "[ ";
writeOperand(PN->getIncomingValue(op), false); Out << ", ";
writeOperand(PN->getIncomingBlock(op), false); Out << " ]";
}
} else if (const ExtractValueInst *EVI = dyn_cast<ExtractValueInst>(&I)) {
Out << ' ';
writeOperand(I.getOperand(0), true);
for (const unsigned *i = EVI->idx_begin(), *e = EVI->idx_end(); i != e; ++i)
Out << ", " << *i;
} else if (const InsertValueInst *IVI = dyn_cast<InsertValueInst>(&I)) {
Out << ' ';
writeOperand(I.getOperand(0), true); Out << ", ";
writeOperand(I.getOperand(1), true);
for (const unsigned *i = IVI->idx_begin(), *e = IVI->idx_end(); i != e; ++i)
Out << ", " << *i;
} else if (const LandingPadInst *LPI = dyn_cast<LandingPadInst>(&I)) {
Out << ' ';
TypePrinter.print(I.getType(), Out);
if (LPI->isCleanup() || LPI->getNumClauses() != 0)
Out << '\n';
if (LPI->isCleanup())
Out << " cleanup";
for (unsigned i = 0, e = LPI->getNumClauses(); i != e; ++i) {
if (i != 0 || LPI->isCleanup()) Out << "\n";
if (LPI->isCatch(i))
Out << " catch ";
else
Out << " filter ";
writeOperand(LPI->getClause(i), true);
}
[IR] Reformulate LLVM's EH funclet IR While we have successfully implemented a funclet-oriented EH scheme on top of LLVM IR, our scheme has some notable deficiencies: - catchendpad and cleanupendpad are necessary in the current design but they are difficult to explain to others, even to seasoned LLVM experts. - catchendpad and cleanupendpad are optimization barriers. They cannot be split and force all potentially throwing call-sites to be invokes. This has a noticable effect on the quality of our code generation. - catchpad, while similar in some aspects to invoke, is fairly awkward. It is unsplittable, starts a funclet, and has control flow to other funclets. - The nesting relationship between funclets is currently a property of control flow edges. Because of this, we are forced to carefully analyze the flow graph to see if there might potentially exist illegal nesting among funclets. While we have logic to clone funclets when they are illegally nested, it would be nicer if we had a representation which forbade them upfront. Let's clean this up a bit by doing the following: - Instead, make catchpad more like cleanuppad and landingpad: no control flow, just a bunch of simple operands; catchpad would be splittable. - Introduce catchswitch, a control flow instruction designed to model the constraints of funclet oriented EH. - Make funclet scoping explicit by having funclet instructions consume the token produced by the funclet which contains them. - Remove catchendpad and cleanupendpad. Their presence can be inferred implicitly using coloring information. N.B. The state numbering code for the CLR has been updated but the veracity of it's output cannot be spoken for. An expert should take a look to make sure the results are reasonable. Reviewers: rnk, JosephTremoulet, andrew.w.kaylor Differential Revision: http://reviews.llvm.org/D15139 llvm-svn: 255422
2015-12-12 06:38:55 +01:00
} else if (const auto *CatchSwitch = dyn_cast<CatchSwitchInst>(&I)) {
Out << " within ";
writeOperand(CatchSwitch->getParentPad(), /*PrintType=*/false);
Out << " [";
[IR] Reformulate LLVM's EH funclet IR While we have successfully implemented a funclet-oriented EH scheme on top of LLVM IR, our scheme has some notable deficiencies: - catchendpad and cleanupendpad are necessary in the current design but they are difficult to explain to others, even to seasoned LLVM experts. - catchendpad and cleanupendpad are optimization barriers. They cannot be split and force all potentially throwing call-sites to be invokes. This has a noticable effect on the quality of our code generation. - catchpad, while similar in some aspects to invoke, is fairly awkward. It is unsplittable, starts a funclet, and has control flow to other funclets. - The nesting relationship between funclets is currently a property of control flow edges. Because of this, we are forced to carefully analyze the flow graph to see if there might potentially exist illegal nesting among funclets. While we have logic to clone funclets when they are illegally nested, it would be nicer if we had a representation which forbade them upfront. Let's clean this up a bit by doing the following: - Instead, make catchpad more like cleanuppad and landingpad: no control flow, just a bunch of simple operands; catchpad would be splittable. - Introduce catchswitch, a control flow instruction designed to model the constraints of funclet oriented EH. - Make funclet scoping explicit by having funclet instructions consume the token produced by the funclet which contains them. - Remove catchendpad and cleanupendpad. Their presence can be inferred implicitly using coloring information. N.B. The state numbering code for the CLR has been updated but the veracity of it's output cannot be spoken for. An expert should take a look to make sure the results are reasonable. Reviewers: rnk, JosephTremoulet, andrew.w.kaylor Differential Revision: http://reviews.llvm.org/D15139 llvm-svn: 255422
2015-12-12 06:38:55 +01:00
unsigned Op = 0;
for (const BasicBlock *PadBB : CatchSwitch->handlers()) {
if (Op > 0)
Out << ", ";
writeOperand(PadBB, /*PrintType=*/true);
++Op;
}
Out << "] unwind ";
if (const BasicBlock *UnwindDest = CatchSwitch->getUnwindDest())
writeOperand(UnwindDest, /*PrintType=*/true);
else
Out << "to caller";
} else if (const auto *FPI = dyn_cast<FuncletPadInst>(&I)) {
Out << " within ";
writeOperand(FPI->getParentPad(), /*PrintType=*/false);
Out << " [";
for (unsigned Op = 0, NumOps = FPI->getNumArgOperands(); Op < NumOps;
++Op) {
if (Op > 0)
Out << ", ";
[IR] Reformulate LLVM's EH funclet IR While we have successfully implemented a funclet-oriented EH scheme on top of LLVM IR, our scheme has some notable deficiencies: - catchendpad and cleanupendpad are necessary in the current design but they are difficult to explain to others, even to seasoned LLVM experts. - catchendpad and cleanupendpad are optimization barriers. They cannot be split and force all potentially throwing call-sites to be invokes. This has a noticable effect on the quality of our code generation. - catchpad, while similar in some aspects to invoke, is fairly awkward. It is unsplittable, starts a funclet, and has control flow to other funclets. - The nesting relationship between funclets is currently a property of control flow edges. Because of this, we are forced to carefully analyze the flow graph to see if there might potentially exist illegal nesting among funclets. While we have logic to clone funclets when they are illegally nested, it would be nicer if we had a representation which forbade them upfront. Let's clean this up a bit by doing the following: - Instead, make catchpad more like cleanuppad and landingpad: no control flow, just a bunch of simple operands; catchpad would be splittable. - Introduce catchswitch, a control flow instruction designed to model the constraints of funclet oriented EH. - Make funclet scoping explicit by having funclet instructions consume the token produced by the funclet which contains them. - Remove catchendpad and cleanupendpad. Their presence can be inferred implicitly using coloring information. N.B. The state numbering code for the CLR has been updated but the veracity of it's output cannot be spoken for. An expert should take a look to make sure the results are reasonable. Reviewers: rnk, JosephTremoulet, andrew.w.kaylor Differential Revision: http://reviews.llvm.org/D15139 llvm-svn: 255422
2015-12-12 06:38:55 +01:00
writeOperand(FPI->getArgOperand(Op), /*PrintType=*/true);
}
[IR] Reformulate LLVM's EH funclet IR While we have successfully implemented a funclet-oriented EH scheme on top of LLVM IR, our scheme has some notable deficiencies: - catchendpad and cleanupendpad are necessary in the current design but they are difficult to explain to others, even to seasoned LLVM experts. - catchendpad and cleanupendpad are optimization barriers. They cannot be split and force all potentially throwing call-sites to be invokes. This has a noticable effect on the quality of our code generation. - catchpad, while similar in some aspects to invoke, is fairly awkward. It is unsplittable, starts a funclet, and has control flow to other funclets. - The nesting relationship between funclets is currently a property of control flow edges. Because of this, we are forced to carefully analyze the flow graph to see if there might potentially exist illegal nesting among funclets. While we have logic to clone funclets when they are illegally nested, it would be nicer if we had a representation which forbade them upfront. Let's clean this up a bit by doing the following: - Instead, make catchpad more like cleanuppad and landingpad: no control flow, just a bunch of simple operands; catchpad would be splittable. - Introduce catchswitch, a control flow instruction designed to model the constraints of funclet oriented EH. - Make funclet scoping explicit by having funclet instructions consume the token produced by the funclet which contains them. - Remove catchendpad and cleanupendpad. Their presence can be inferred implicitly using coloring information. N.B. The state numbering code for the CLR has been updated but the veracity of it's output cannot be spoken for. An expert should take a look to make sure the results are reasonable. Reviewers: rnk, JosephTremoulet, andrew.w.kaylor Differential Revision: http://reviews.llvm.org/D15139 llvm-svn: 255422
2015-12-12 06:38:55 +01:00
Out << ']';
} else if (isa<ReturnInst>(I) && !Operand) {
Out << " void";
} else if (const auto *CRI = dyn_cast<CatchReturnInst>(&I)) {
[IR] Reformulate LLVM's EH funclet IR While we have successfully implemented a funclet-oriented EH scheme on top of LLVM IR, our scheme has some notable deficiencies: - catchendpad and cleanupendpad are necessary in the current design but they are difficult to explain to others, even to seasoned LLVM experts. - catchendpad and cleanupendpad are optimization barriers. They cannot be split and force all potentially throwing call-sites to be invokes. This has a noticable effect on the quality of our code generation. - catchpad, while similar in some aspects to invoke, is fairly awkward. It is unsplittable, starts a funclet, and has control flow to other funclets. - The nesting relationship between funclets is currently a property of control flow edges. Because of this, we are forced to carefully analyze the flow graph to see if there might potentially exist illegal nesting among funclets. While we have logic to clone funclets when they are illegally nested, it would be nicer if we had a representation which forbade them upfront. Let's clean this up a bit by doing the following: - Instead, make catchpad more like cleanuppad and landingpad: no control flow, just a bunch of simple operands; catchpad would be splittable. - Introduce catchswitch, a control flow instruction designed to model the constraints of funclet oriented EH. - Make funclet scoping explicit by having funclet instructions consume the token produced by the funclet which contains them. - Remove catchendpad and cleanupendpad. Their presence can be inferred implicitly using coloring information. N.B. The state numbering code for the CLR has been updated but the veracity of it's output cannot be spoken for. An expert should take a look to make sure the results are reasonable. Reviewers: rnk, JosephTremoulet, andrew.w.kaylor Differential Revision: http://reviews.llvm.org/D15139 llvm-svn: 255422
2015-12-12 06:38:55 +01:00
Out << " from ";
writeOperand(CRI->getOperand(0), /*PrintType=*/false);
Out << " to ";
[IR] Reformulate LLVM's EH funclet IR While we have successfully implemented a funclet-oriented EH scheme on top of LLVM IR, our scheme has some notable deficiencies: - catchendpad and cleanupendpad are necessary in the current design but they are difficult to explain to others, even to seasoned LLVM experts. - catchendpad and cleanupendpad are optimization barriers. They cannot be split and force all potentially throwing call-sites to be invokes. This has a noticable effect on the quality of our code generation. - catchpad, while similar in some aspects to invoke, is fairly awkward. It is unsplittable, starts a funclet, and has control flow to other funclets. - The nesting relationship between funclets is currently a property of control flow edges. Because of this, we are forced to carefully analyze the flow graph to see if there might potentially exist illegal nesting among funclets. While we have logic to clone funclets when they are illegally nested, it would be nicer if we had a representation which forbade them upfront. Let's clean this up a bit by doing the following: - Instead, make catchpad more like cleanuppad and landingpad: no control flow, just a bunch of simple operands; catchpad would be splittable. - Introduce catchswitch, a control flow instruction designed to model the constraints of funclet oriented EH. - Make funclet scoping explicit by having funclet instructions consume the token produced by the funclet which contains them. - Remove catchendpad and cleanupendpad. Their presence can be inferred implicitly using coloring information. N.B. The state numbering code for the CLR has been updated but the veracity of it's output cannot be spoken for. An expert should take a look to make sure the results are reasonable. Reviewers: rnk, JosephTremoulet, andrew.w.kaylor Differential Revision: http://reviews.llvm.org/D15139 llvm-svn: 255422
2015-12-12 06:38:55 +01:00
writeOperand(CRI->getOperand(1), /*PrintType=*/true);
} else if (const auto *CRI = dyn_cast<CleanupReturnInst>(&I)) {
[IR] Reformulate LLVM's EH funclet IR While we have successfully implemented a funclet-oriented EH scheme on top of LLVM IR, our scheme has some notable deficiencies: - catchendpad and cleanupendpad are necessary in the current design but they are difficult to explain to others, even to seasoned LLVM experts. - catchendpad and cleanupendpad are optimization barriers. They cannot be split and force all potentially throwing call-sites to be invokes. This has a noticable effect on the quality of our code generation. - catchpad, while similar in some aspects to invoke, is fairly awkward. It is unsplittable, starts a funclet, and has control flow to other funclets. - The nesting relationship between funclets is currently a property of control flow edges. Because of this, we are forced to carefully analyze the flow graph to see if there might potentially exist illegal nesting among funclets. While we have logic to clone funclets when they are illegally nested, it would be nicer if we had a representation which forbade them upfront. Let's clean this up a bit by doing the following: - Instead, make catchpad more like cleanuppad and landingpad: no control flow, just a bunch of simple operands; catchpad would be splittable. - Introduce catchswitch, a control flow instruction designed to model the constraints of funclet oriented EH. - Make funclet scoping explicit by having funclet instructions consume the token produced by the funclet which contains them. - Remove catchendpad and cleanupendpad. Their presence can be inferred implicitly using coloring information. N.B. The state numbering code for the CLR has been updated but the veracity of it's output cannot be spoken for. An expert should take a look to make sure the results are reasonable. Reviewers: rnk, JosephTremoulet, andrew.w.kaylor Differential Revision: http://reviews.llvm.org/D15139 llvm-svn: 255422
2015-12-12 06:38:55 +01:00
Out << " from ";
writeOperand(CRI->getOperand(0), /*PrintType=*/false);
Out << " unwind ";
if (CRI->hasUnwindDest())
[IR] Reformulate LLVM's EH funclet IR While we have successfully implemented a funclet-oriented EH scheme on top of LLVM IR, our scheme has some notable deficiencies: - catchendpad and cleanupendpad are necessary in the current design but they are difficult to explain to others, even to seasoned LLVM experts. - catchendpad and cleanupendpad are optimization barriers. They cannot be split and force all potentially throwing call-sites to be invokes. This has a noticable effect on the quality of our code generation. - catchpad, while similar in some aspects to invoke, is fairly awkward. It is unsplittable, starts a funclet, and has control flow to other funclets. - The nesting relationship between funclets is currently a property of control flow edges. Because of this, we are forced to carefully analyze the flow graph to see if there might potentially exist illegal nesting among funclets. While we have logic to clone funclets when they are illegally nested, it would be nicer if we had a representation which forbade them upfront. Let's clean this up a bit by doing the following: - Instead, make catchpad more like cleanuppad and landingpad: no control flow, just a bunch of simple operands; catchpad would be splittable. - Introduce catchswitch, a control flow instruction designed to model the constraints of funclet oriented EH. - Make funclet scoping explicit by having funclet instructions consume the token produced by the funclet which contains them. - Remove catchendpad and cleanupendpad. Their presence can be inferred implicitly using coloring information. N.B. The state numbering code for the CLR has been updated but the veracity of it's output cannot be spoken for. An expert should take a look to make sure the results are reasonable. Reviewers: rnk, JosephTremoulet, andrew.w.kaylor Differential Revision: http://reviews.llvm.org/D15139 llvm-svn: 255422
2015-12-12 06:38:55 +01:00
writeOperand(CRI->getOperand(1), /*PrintType=*/true);
else
Out << "to caller";
} else if (const CallInst *CI = dyn_cast<CallInst>(&I)) {
// Print the calling convention being used.
if (CI->getCallingConv() != CallingConv::C) {
Out << " ";
PrintCallingConv(CI->getCallingConv(), Out);
}
Operand = CI->getCalledValue();
FunctionType *FTy = CI->getFunctionType();
Type *RetTy = FTy->getReturnType();
const AttributeList &PAL = CI->getAttributes();
if (PAL.hasAttributes(AttributeList::ReturnIndex))
Out << ' ' << PAL.getAsString(AttributeList::ReturnIndex);
// If possible, print out the short form of the call instruction. We can
// only do this if the first argument is a pointer to a nonvararg function,
// and if the return type is not a pointer to a function.
//
Out << ' ';
[opaque pointer type] Add textual IR support for explicit type parameter to the call instruction See r230786 and r230794 for similar changes to gep and load respectively. Call is a bit different because it often doesn't have a single explicit type - usually the type is deduced from the arguments, and just the return type is explicit. In those cases there's no need to change the IR. When that's not the case, the IR usually contains the pointer type of the first operand - but since typed pointers are going away, that representation is insufficient so I'm just stripping the "pointerness" of the explicit type away. This does make the IR a bit weird - it /sort of/ reads like the type of the first operand: "call void () %x(" but %x is actually of type "void ()*" and will eventually be just of type "ptr". But this seems not too bad and I don't think it would benefit from repeating the type ("void (), void () * %x(" and then eventually "void (), ptr %x(") as has been done with gep and load. This also has a side benefit: since the explicit type is no longer a pointer, there's no ambiguity between an explicit type and a function that returns a function pointer. Previously this case needed an explicit type (eg: a function returning a void() function was written as "call void () () * @x(" rather than "call void () * @x(" because of the ambiguity between a function returning a pointer to a void() function and a function returning void). No ambiguity means even function pointer return types can just be written alone, without writing the whole function's type. This leaves /only/ the varargs case where the explicit type is required. Given the special type syntax in call instructions, the regex-fu used for migration was a bit more involved in its own unique way (as every one of these is) so here it is. Use it in conjunction with the apply.sh script and associated find/xargs commands I've provided in rr230786 to migrate your out of tree tests. Do let me know if any of this doesn't cover your cases & we can iterate on a more general script/regexes to help others with out of tree tests. About 9 test cases couldn't be automatically migrated - half of those were functions returning function pointers, where I just had to manually delete the function argument types now that we didn't need an explicit function type there. The other half were typedefs of function types used in calls - just had to manually drop the * from those. import fileinput import sys import re pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)') addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$") func_end = re.compile("(?:void.*|\)\s*)\*$") def conv(match, line): if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)): return line return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():] for line in sys.stdin: sys.stdout.write(conv(re.search(pat, line), line)) llvm-svn: 235145
2015-04-17 01:24:18 +02:00
TypePrinter.print(FTy->isVarArg() ? FTy : RetTy, Out);
Out << ' ';
writeOperand(Operand, false);
Out << '(';
for (unsigned op = 0, Eop = CI->getNumArgOperands(); op < Eop; ++op) {
if (op > 0)
Out << ", ";
writeParamOperand(CI->getArgOperand(op), PAL.getParamAttributes(op));
2001-06-06 22:29:01 +02:00
}
// Emit an ellipsis if this is a musttail call in a vararg function. This
// is only to aid readability, musttail calls forward varargs by default.
if (CI->isMustTailCall() && CI->getParent() &&
CI->getParent()->getParent() &&
CI->getParent()->getParent()->isVarArg())
Out << ", ...";
Out << ')';
if (PAL.hasAttributes(AttributeList::FunctionIndex))
Out << " #" << Machine.getAttributeGroupSlot(PAL.getFnAttributes());
writeOperandBundles(CI);
} else if (const InvokeInst *II = dyn_cast<InvokeInst>(&I)) {
Operand = II->getCalledValue();
FunctionType *FTy = II->getFunctionType();
Type *RetTy = FTy->getReturnType();
const AttributeList &PAL = II->getAttributes();
// Print the calling convention being used.
if (II->getCallingConv() != CallingConv::C) {
Out << " ";
PrintCallingConv(II->getCallingConv(), Out);
}
if (PAL.hasAttributes(AttributeList::ReturnIndex))
Out << ' ' << PAL.getAsString(AttributeList::ReturnIndex);
// If possible, print out the short form of the invoke instruction. We can
// only do this if the first argument is a pointer to a nonvararg function,
// and if the return type is not a pointer to a function.
//
Out << ' ';
TypePrinter.print(FTy->isVarArg() ? FTy : RetTy, Out);
Out << ' ';
writeOperand(Operand, false);
Out << '(';
for (unsigned op = 0, Eop = II->getNumArgOperands(); op < Eop; ++op) {
if (op)
Out << ", ";
writeParamOperand(II->getArgOperand(op), PAL.getParamAttributes(op));
}
Out << ')';
if (PAL.hasAttributes(AttributeList::FunctionIndex))
Out << " #" << Machine.getAttributeGroupSlot(PAL.getFnAttributes());
writeOperandBundles(II);
Out << "\n to ";
writeOperand(II->getNormalDest(), true);
Out << " unwind ";
writeOperand(II->getUnwindDest(), true);
} else if (const AllocaInst *AI = dyn_cast<AllocaInst>(&I)) {
Out << ' ';
if (AI->isUsedWithInAlloca())
Out << "inalloca ";
if (AI->isSwiftError())
Out << "swifterror ";
TypePrinter.print(AI->getAllocatedType(), Out);
// Explicitly write the array size if the code is broken, if it's an array
// allocation, or if the type is not canonical for scalar allocations. The
// latter case prevents the type from mutating when round-tripping through
// assembly.
if (!AI->getArraySize() || AI->isArrayAllocation() ||
!AI->getArraySize()->getType()->isIntegerTy(32)) {
Out << ", ";
writeOperand(AI->getArraySize(), true);
2001-06-06 22:29:01 +02:00
}
if (AI->getAlignment()) {
Out << ", align " << AI->getAlignment();
}
unsigned AddrSpace = AI->getType()->getAddressSpace();
if (AddrSpace != 0) {
Out << ", addrspace(" << AddrSpace << ')';
}
} else if (isa<CastInst>(I)) {
if (Operand) {
Out << ' ';
writeOperand(Operand, true); // Work with broken code
}
Out << " to ";
TypePrinter.print(I.getType(), Out);
} else if (isa<VAArgInst>(I)) {
if (Operand) {
Out << ' ';
writeOperand(Operand, true); // Work with broken code
}
Out << ", ";
TypePrinter.print(I.getType(), Out);
} else if (Operand) { // Print the normal way.
if (const auto *GEP = dyn_cast<GetElementPtrInst>(&I)) {
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-27 20:29:02 +01:00
Out << ' ';
TypePrinter.print(GEP->getSourceElementType(), Out);
Out << ',';
} else if (const auto *LI = dyn_cast<LoadInst>(&I)) {
Out << ' ';
TypePrinter.print(LI->getType(), Out);
Out << ',';
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-27 20:29:02 +01:00
}
2001-06-06 22:29:01 +02:00
// PrintAllTypes - Instructions who have operands of all the same type
2001-06-06 22:29:01 +02:00
// omit the type from all but the first operand. If the instruction has
// different type operands (for example br), then they are all printed.
bool PrintAllTypes = false;
Type *TheType = Operand->getType();
2001-06-06 22:29:01 +02:00
// Select, Store and ShuffleVector always print all types.
if (isa<SelectInst>(I) || isa<StoreInst>(I) || isa<ShuffleVectorInst>(I)
|| isa<ReturnInst>(I)) {
PrintAllTypes = true;
} else {
for (unsigned i = 1, E = I.getNumOperands(); i != E; ++i) {
Operand = I.getOperand(i);
// note that Operand shouldn't be null, but the test helps make dump()
// more tolerant of malformed IR
if (Operand && Operand->getType() != TheType) {
PrintAllTypes = true; // We have differing types! Print them all!
break;
}
2001-06-06 22:29:01 +02:00
}
}
if (!PrintAllTypes) {
Out << ' ';
TypePrinter.print(TheType, Out);
}
2001-06-06 22:29:01 +02:00
Out << ' ';
for (unsigned i = 0, E = I.getNumOperands(); i != E; ++i) {
if (i) Out << ", ";
writeOperand(I.getOperand(i), PrintAllTypes);
2001-06-06 22:29:01 +02:00
}
}
// Print atomic ordering/alignment for memory operations
if (const LoadInst *LI = dyn_cast<LoadInst>(&I)) {
if (LI->isAtomic())
writeAtomic(LI->getContext(), LI->getOrdering(), LI->getSyncScopeID());
if (LI->getAlignment())
Out << ", align " << LI->getAlignment();
} else if (const StoreInst *SI = dyn_cast<StoreInst>(&I)) {
if (SI->isAtomic())
writeAtomic(SI->getContext(), SI->getOrdering(), SI->getSyncScopeID());
if (SI->getAlignment())
Out << ", align " << SI->getAlignment();
} else if (const AtomicCmpXchgInst *CXI = dyn_cast<AtomicCmpXchgInst>(&I)) {
writeAtomicCmpXchg(CXI->getContext(), CXI->getSuccessOrdering(),
CXI->getFailureOrdering(), CXI->getSyncScopeID());
} else if (const AtomicRMWInst *RMWI = dyn_cast<AtomicRMWInst>(&I)) {
writeAtomic(RMWI->getContext(), RMWI->getOrdering(),
RMWI->getSyncScopeID());
} else if (const FenceInst *FI = dyn_cast<FenceInst>(&I)) {
writeAtomic(FI->getContext(), FI->getOrdering(), FI->getSyncScopeID());
}
2001-06-06 22:29:01 +02:00
// Print Metadata info.
SmallVector<std::pair<unsigned, MDNode *>, 4> InstMD;
I.getAllMetadata(InstMD);
printMetadataAttachments(InstMD, ", ");
// Print a nice comment.
printInfoComment(I);
}
void AssemblyWriter::printMetadataAttachments(
const SmallVectorImpl<std::pair<unsigned, MDNode *>> &MDs,
StringRef Separator) {
if (MDs.empty())
return;
if (MDNames.empty())
MDs[0].second->getContext().getMDKindNames(MDNames);
for (const auto &I : MDs) {
unsigned Kind = I.first;
Out << Separator;
if (Kind < MDNames.size()) {
Out << "!";
printMetadataIdentifier(MDNames[Kind], Out);
} else
Out << "!<unknown kind #" << Kind << ">";
Out << ' ';
WriteAsOperandInternal(Out, I.second, &TypePrinter, &Machine, TheModule);
}
2001-06-06 22:29:01 +02:00
}
void AssemblyWriter::writeMDNode(unsigned Slot, const MDNode *Node) {
IR: Make metadata typeless in assembly Now that `Metadata` is typeless, reflect that in the assembly. These are the matching assembly changes for the metadata/value split in r223802. - Only use the `metadata` type when referencing metadata from a call intrinsic -- i.e., only when it's used as a `Value`. - Stop pretending that `ValueAsMetadata` is wrapped in an `MDNode` when referencing it from call intrinsics. So, assembly like this: define @foo(i32 %v) { call void @llvm.foo(metadata !{i32 %v}, metadata !0) call void @llvm.foo(metadata !{i32 7}, metadata !0) call void @llvm.foo(metadata !1, metadata !0) call void @llvm.foo(metadata !3, metadata !0) call void @llvm.foo(metadata !{metadata !3}, metadata !0) ret void, !bar !2 } !0 = metadata !{metadata !2} !1 = metadata !{i32* @global} !2 = metadata !{metadata !3} !3 = metadata !{} turns into this: define @foo(i32 %v) { call void @llvm.foo(metadata i32 %v, metadata !0) call void @llvm.foo(metadata i32 7, metadata !0) call void @llvm.foo(metadata i32* @global, metadata !0) call void @llvm.foo(metadata !3, metadata !0) call void @llvm.foo(metadata !{!3}, metadata !0) ret void, !bar !2 } !0 = !{!2} !1 = !{i32* @global} !2 = !{!3} !3 = !{} I wrote an upgrade script that handled almost all of the tests in llvm and many of the tests in cfe (even handling many `CHECK` lines). I've attached it (or will attach it in a moment if you're speedy) to PR21532 to help everyone update their out-of-tree testcases. This is part of PR21532. llvm-svn: 224257
2014-12-15 20:07:53 +01:00
Out << '!' << Slot << " = ";
printMDNodeBody(Node);
Out << "\n";
}
void AssemblyWriter::writeAllMDNodes() {
SmallVector<const MDNode *, 16> Nodes;
Nodes.resize(Machine.mdn_size());
for (SlotTracker::mdn_iterator I = Machine.mdn_begin(), E = Machine.mdn_end();
I != E; ++I)
Nodes[I->second] = cast<MDNode>(I->first);
2011-09-30 21:48:58 +02:00
for (unsigned i = 0, e = Nodes.size(); i != e; ++i) {
writeMDNode(i, Nodes[i]);
}
}
void AssemblyWriter::printMDNodeBody(const MDNode *Node) {
WriteMDNodeBodyInternal(Out, Node, &TypePrinter, &Machine, TheModule);
}
2001-06-06 22:29:01 +02:00
void AssemblyWriter::writeAllAttributeGroups() {
std::vector<std::pair<AttributeSet, unsigned>> asVec;
asVec.resize(Machine.as_size());
for (SlotTracker::as_iterator I = Machine.as_begin(), E = Machine.as_end();
I != E; ++I)
asVec[I->second] = *I;
for (const auto &I : asVec)
Out << "attributes #" << I.second << " = { "
<< I.first.getAsString(true) << " }\n";
}
void AssemblyWriter::printUseListOrder(const UseListOrder &Order) {
bool IsInFunction = Machine.getFunction();
if (IsInFunction)
Out << " ";
Out << "uselistorder";
if (const BasicBlock *BB =
IsInFunction ? nullptr : dyn_cast<BasicBlock>(Order.V)) {
Out << "_bb ";
writeOperand(BB->getParent(), false);
Out << ", ";
writeOperand(BB, false);
} else {
Out << " ";
writeOperand(Order.V, true);
}
Out << ", { ";
assert(Order.Shuffle.size() >= 2 && "Shuffle too small");
Out << Order.Shuffle[0];
for (unsigned I = 1, E = Order.Shuffle.size(); I != E; ++I)
Out << ", " << Order.Shuffle[I];
Out << " }\n";
}
void AssemblyWriter::printUseLists(const Function *F) {
auto hasMore =
[&]() { return !UseListOrders.empty() && UseListOrders.back().F == F; };
if (!hasMore())
// Nothing to do.
return;
Out << "\n; uselistorder directives\n";
while (hasMore()) {
printUseListOrder(UseListOrders.back());
UseListOrders.pop_back();
}
}
2001-06-06 22:29:01 +02:00
//===----------------------------------------------------------------------===//
// External Interface declarations
//===----------------------------------------------------------------------===//
void Function::print(raw_ostream &ROS, AssemblyAnnotationWriter *AAW,
bool ShouldPreserveUseListOrder,
bool IsForDebug) const {
SlotTracker SlotTable(this->getParent());
formatted_raw_ostream OS(ROS);
AssemblyWriter W(OS, SlotTable, this->getParent(), AAW,
IsForDebug,
ShouldPreserveUseListOrder);
W.printFunction(this);
}
void Module::print(raw_ostream &ROS, AssemblyAnnotationWriter *AAW,
bool ShouldPreserveUseListOrder, bool IsForDebug) const {
SlotTracker SlotTable(this);
formatted_raw_ostream OS(ROS);
AssemblyWriter W(OS, SlotTable, this, AAW, IsForDebug,
ShouldPreserveUseListOrder);
W.printModule(this);
}
2001-06-06 22:29:01 +02:00
void NamedMDNode::print(raw_ostream &ROS, bool IsForDebug) const {
SlotTracker SlotTable(getParent());
formatted_raw_ostream OS(ROS);
AssemblyWriter W(OS, SlotTable, getParent(), nullptr, IsForDebug);
W.printNamedMDNode(this);
}
void NamedMDNode::print(raw_ostream &ROS, ModuleSlotTracker &MST,
bool IsForDebug) const {
Optional<SlotTracker> LocalST;
SlotTracker *SlotTable;
if (auto *ST = MST.getMachine())
SlotTable = ST;
else {
LocalST.emplace(getParent());
SlotTable = &*LocalST;
}
formatted_raw_ostream OS(ROS);
AssemblyWriter W(OS, *SlotTable, getParent(), nullptr, IsForDebug);
W.printNamedMDNode(this);
}
void Comdat::print(raw_ostream &ROS, bool /*IsForDebug*/) const {
PrintLLVMName(ROS, getName(), ComdatPrefix);
ROS << " = comdat ";
switch (getSelectionKind()) {
case Comdat::Any:
ROS << "any";
break;
case Comdat::ExactMatch:
ROS << "exactmatch";
break;
case Comdat::Largest:
ROS << "largest";
break;
case Comdat::NoDuplicates:
ROS << "noduplicates";
break;
case Comdat::SameSize:
ROS << "samesize";
break;
}
ROS << '\n';
}
void Type::print(raw_ostream &OS, bool /*IsForDebug*/, bool NoDetails) const {
TypePrinting TP;
TP.print(const_cast<Type*>(this), OS);
2011-09-30 21:48:58 +02:00
if (NoDetails)
return;
// If the type is a named struct type, print the body as well.
if (StructType *STy = dyn_cast<StructType>(const_cast<Type*>(this)))
if (!STy->isLiteral()) {
OS << " = type ";
TP.printStructBody(STy, OS);
}
2001-06-06 22:29:01 +02:00
}
IR: Make Metadata::print() reliable and useful Replumb the `AsmWriter` so that `Metadata::print()` is generally useful. (Similarly change `Metadata::printAsOperand()`.) - `SlotTracker` now has a mode where all metadata will be correctly numbered when initializing a `Module`. Normally, `Metadata` only referenced from within `Function`s gets numbered when the `Function` is incorporated. - `Metadata::print()` and `Metadata::printAsOperand()` (and `Metadata::dump()`) now take an optional `Module` argument. When provided, `SlotTracker` is initialized with the new mode, and the numbering will be complete and consistent for all calls to `print()`. - `Value::print()` uses the new `SlotTracker` mode when printing intrinsics with `MDNode` operands, `MetadataAsValue` operands, or the bodies of functions. Thus, metadata numbering will be consistent between calls to `Metadata::print()` and `Value::print()`. - `Metadata::print()` (and `Metadata::dump()`) now print the full definition of `MDNode`s: !5 = !{!6, !"abc", !7} This matches behaviour for `Value::print()`, which includes the name of instructions. - Updated call sites in `Verifier` to call `print()` instead of `printAsOperand()`. All this, so that `Verifier` can print out useful failure messages that involve `Metadata` for PR22777. Note that `Metadata::printAsOperand()` previously took an optional `bool` and `Module` operand. The former was cargo-culted from `Value::printAsOperand()` and wasn't doing anything useful. The latter didn't give consistent results (without the new `SlotTracker` mode). llvm-svn: 232275
2015-03-14 21:19:36 +01:00
static bool isReferencingMDNode(const Instruction &I) {
if (const auto *CI = dyn_cast<CallInst>(&I))
if (Function *F = CI->getCalledFunction())
if (F->isIntrinsic())
IR: Make Metadata::print() reliable and useful Replumb the `AsmWriter` so that `Metadata::print()` is generally useful. (Similarly change `Metadata::printAsOperand()`.) - `SlotTracker` now has a mode where all metadata will be correctly numbered when initializing a `Module`. Normally, `Metadata` only referenced from within `Function`s gets numbered when the `Function` is incorporated. - `Metadata::print()` and `Metadata::printAsOperand()` (and `Metadata::dump()`) now take an optional `Module` argument. When provided, `SlotTracker` is initialized with the new mode, and the numbering will be complete and consistent for all calls to `print()`. - `Value::print()` uses the new `SlotTracker` mode when printing intrinsics with `MDNode` operands, `MetadataAsValue` operands, or the bodies of functions. Thus, metadata numbering will be consistent between calls to `Metadata::print()` and `Value::print()`. - `Metadata::print()` (and `Metadata::dump()`) now print the full definition of `MDNode`s: !5 = !{!6, !"abc", !7} This matches behaviour for `Value::print()`, which includes the name of instructions. - Updated call sites in `Verifier` to call `print()` instead of `printAsOperand()`. All this, so that `Verifier` can print out useful failure messages that involve `Metadata` for PR22777. Note that `Metadata::printAsOperand()` previously took an optional `bool` and `Module` operand. The former was cargo-culted from `Value::printAsOperand()` and wasn't doing anything useful. The latter didn't give consistent results (without the new `SlotTracker` mode). llvm-svn: 232275
2015-03-14 21:19:36 +01:00
for (auto &Op : I.operands())
if (auto *V = dyn_cast_or_null<MetadataAsValue>(Op))
if (isa<MDNode>(V->getMetadata()))
return true;
return false;
}
void Value::print(raw_ostream &ROS, bool IsForDebug) const {
bool ShouldInitializeAllMetadata = false;
if (auto *I = dyn_cast<Instruction>(this))
ShouldInitializeAllMetadata = isReferencingMDNode(*I);
else if (isa<Function>(this) || isa<MetadataAsValue>(this))
ShouldInitializeAllMetadata = true;
ModuleSlotTracker MST(getModuleFromVal(this), ShouldInitializeAllMetadata);
print(ROS, MST, IsForDebug);
}
void Value::print(raw_ostream &ROS, ModuleSlotTracker &MST,
bool IsForDebug) const {
formatted_raw_ostream OS(ROS);
SlotTracker EmptySlotTable(static_cast<const Module *>(nullptr));
SlotTracker &SlotTable =
MST.getMachine() ? *MST.getMachine() : EmptySlotTable;
auto incorporateFunction = [&](const Function *F) {
if (F)
MST.incorporateFunction(*F);
};
if (const Instruction *I = dyn_cast<Instruction>(this)) {
incorporateFunction(I->getParent() ? I->getParent()->getParent() : nullptr);
AssemblyWriter W(OS, SlotTable, getModuleFromVal(I), nullptr, IsForDebug);
W.printInstruction(*I);
} else if (const BasicBlock *BB = dyn_cast<BasicBlock>(this)) {
incorporateFunction(BB->getParent());
AssemblyWriter W(OS, SlotTable, getModuleFromVal(BB), nullptr, IsForDebug);
W.printBasicBlock(BB);
} else if (const GlobalValue *GV = dyn_cast<GlobalValue>(this)) {
AssemblyWriter W(OS, SlotTable, GV->getParent(), nullptr, IsForDebug);
if (const GlobalVariable *V = dyn_cast<GlobalVariable>(GV))
W.printGlobal(V);
else if (const Function *F = dyn_cast<Function>(GV))
W.printFunction(F);
else
W.printIndirectSymbol(cast<GlobalIndirectSymbol>(GV));
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
} else if (const MetadataAsValue *V = dyn_cast<MetadataAsValue>(this)) {
V->getMetadata()->print(ROS, MST, getModuleFromVal(V));
} else if (const Constant *C = dyn_cast<Constant>(this)) {
TypePrinting TypePrinter;
TypePrinter.print(C->getType(), OS);
2009-02-28 22:11:05 +01:00
OS << ' ';
WriteConstantInternal(OS, C, TypePrinter, MST.getMachine(), nullptr);
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
} else if (isa<InlineAsm>(this) || isa<Argument>(this)) {
this->printAsOperand(OS, /* PrintType */ true, MST);
} else {
llvm_unreachable("Unknown value to print out!");
}
}
/// Print without a type, skipping the TypePrinting object.
///
/// \return \c true iff printing was successful.
static bool printWithoutType(const Value &V, raw_ostream &O,
SlotTracker *Machine, const Module *M) {
if (V.hasName() || isa<GlobalValue>(V) ||
(!isa<Constant>(V) && !isa<MetadataAsValue>(V))) {
WriteAsOperandInternal(O, &V, nullptr, Machine, M);
return true;
}
return false;
}
static void printAsOperandImpl(const Value &V, raw_ostream &O, bool PrintType,
ModuleSlotTracker &MST) {
[MIR] Making MIR Printing, opt -dot-cfg, and -debug printing faster Value::printAsOperand has been scanning the entire module just to print a single value as an operand, regardless being asked to print a type or not at all, and regardless really needing to scan the module to print a type. It made some of the users of the method exceptionally slow on large IR-modules (or large MIR-files with large IR-modules embedded). This patch defers scanning a module looking for struct types, mostly numbered struct types, as much as possible, speeding up those users w/o changing any APIs at all. See speedup examples below: Release Build: # 83 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 133 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output Release + Asserts Build: # 95 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 146 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output # 1096 seconds -> 553 seconds time ./bin/llc -debug-only=isel -fast-isel=false -stop-after=isel \ sqlite3.O0.ll -o /dev/null 2> err where sqlite3.O0.ll is non-optimized IR produced from sqlite-amalgamation (http://sqlite.org/download.html), which is entire SQLite3 implementation in a single C-file. Benchmarked on 4-cores / 8 threads PCI-E SSD iMac running macOS Reviewers: dexonsmith, bkramer, void, chandlerc, aditya_nandakumar, dsanders, qcolombet, Reviewed By: bogner Subscribers: thegameg, llvm-commits Differential Revision: https://reviews.llvm.org/D44132 llvm-svn: 328246
2018-03-22 22:29:07 +01:00
TypePrinting TypePrinter(MST.getModule());
if (PrintType) {
TypePrinter.print(V.getType(), O);
O << ' ';
}
WriteAsOperandInternal(O, &V, &TypePrinter, MST.getMachine(),
MST.getModule());
}
void Value::printAsOperand(raw_ostream &O, bool PrintType,
const Module *M) const {
if (!M)
M = getModuleFromVal(this);
if (!PrintType)
if (printWithoutType(*this, O, nullptr, M))
return;
SlotTracker Machine(
M, /* ShouldInitializeAllMetadata */ isa<MetadataAsValue>(this));
ModuleSlotTracker MST(Machine, M);
printAsOperandImpl(*this, O, PrintType, MST);
}
void Value::printAsOperand(raw_ostream &O, bool PrintType,
ModuleSlotTracker &MST) const {
if (!PrintType)
if (printWithoutType(*this, O, MST.getMachine(), MST.getModule()))
return;
printAsOperandImpl(*this, O, PrintType, MST);
}
IR: Make Metadata::print() reliable and useful Replumb the `AsmWriter` so that `Metadata::print()` is generally useful. (Similarly change `Metadata::printAsOperand()`.) - `SlotTracker` now has a mode where all metadata will be correctly numbered when initializing a `Module`. Normally, `Metadata` only referenced from within `Function`s gets numbered when the `Function` is incorporated. - `Metadata::print()` and `Metadata::printAsOperand()` (and `Metadata::dump()`) now take an optional `Module` argument. When provided, `SlotTracker` is initialized with the new mode, and the numbering will be complete and consistent for all calls to `print()`. - `Value::print()` uses the new `SlotTracker` mode when printing intrinsics with `MDNode` operands, `MetadataAsValue` operands, or the bodies of functions. Thus, metadata numbering will be consistent between calls to `Metadata::print()` and `Value::print()`. - `Metadata::print()` (and `Metadata::dump()`) now print the full definition of `MDNode`s: !5 = !{!6, !"abc", !7} This matches behaviour for `Value::print()`, which includes the name of instructions. - Updated call sites in `Verifier` to call `print()` instead of `printAsOperand()`. All this, so that `Verifier` can print out useful failure messages that involve `Metadata` for PR22777. Note that `Metadata::printAsOperand()` previously took an optional `bool` and `Module` operand. The former was cargo-culted from `Value::printAsOperand()` and wasn't doing anything useful. The latter didn't give consistent results (without the new `SlotTracker` mode). llvm-svn: 232275
2015-03-14 21:19:36 +01:00
static void printMetadataImpl(raw_ostream &ROS, const Metadata &MD,
ModuleSlotTracker &MST, const Module *M,
bool OnlyAsOperand) {
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
formatted_raw_ostream OS(ROS);
[MIR] Making MIR Printing, opt -dot-cfg, and -debug printing faster Value::printAsOperand has been scanning the entire module just to print a single value as an operand, regardless being asked to print a type or not at all, and regardless really needing to scan the module to print a type. It made some of the users of the method exceptionally slow on large IR-modules (or large MIR-files with large IR-modules embedded). This patch defers scanning a module looking for struct types, mostly numbered struct types, as much as possible, speeding up those users w/o changing any APIs at all. See speedup examples below: Release Build: # 83 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 133 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output Release + Asserts Build: # 95 seconds -> 5.5 seconds time ./bin/llc -start-before=irtranslator -stop-after=irtranslator \ -global-isel -global-isel-abort=2 -simplify-mir sqlite3.O0.ll -o \ sqlite3.O0.ll.regbankselected.mir # 146 seconds -> 6.2 seconds time ./bin/opt sqlite3.O0.ll -dot-cfg -disable-output # 1096 seconds -> 553 seconds time ./bin/llc -debug-only=isel -fast-isel=false -stop-after=isel \ sqlite3.O0.ll -o /dev/null 2> err where sqlite3.O0.ll is non-optimized IR produced from sqlite-amalgamation (http://sqlite.org/download.html), which is entire SQLite3 implementation in a single C-file. Benchmarked on 4-cores / 8 threads PCI-E SSD iMac running macOS Reviewers: dexonsmith, bkramer, void, chandlerc, aditya_nandakumar, dsanders, qcolombet, Reviewed By: bogner Subscribers: thegameg, llvm-commits Differential Revision: https://reviews.llvm.org/D44132 llvm-svn: 328246
2018-03-22 22:29:07 +01:00
TypePrinting TypePrinter(M);
IR: Make Metadata::print() reliable and useful Replumb the `AsmWriter` so that `Metadata::print()` is generally useful. (Similarly change `Metadata::printAsOperand()`.) - `SlotTracker` now has a mode where all metadata will be correctly numbered when initializing a `Module`. Normally, `Metadata` only referenced from within `Function`s gets numbered when the `Function` is incorporated. - `Metadata::print()` and `Metadata::printAsOperand()` (and `Metadata::dump()`) now take an optional `Module` argument. When provided, `SlotTracker` is initialized with the new mode, and the numbering will be complete and consistent for all calls to `print()`. - `Value::print()` uses the new `SlotTracker` mode when printing intrinsics with `MDNode` operands, `MetadataAsValue` operands, or the bodies of functions. Thus, metadata numbering will be consistent between calls to `Metadata::print()` and `Value::print()`. - `Metadata::print()` (and `Metadata::dump()`) now print the full definition of `MDNode`s: !5 = !{!6, !"abc", !7} This matches behaviour for `Value::print()`, which includes the name of instructions. - Updated call sites in `Verifier` to call `print()` instead of `printAsOperand()`. All this, so that `Verifier` can print out useful failure messages that involve `Metadata` for PR22777. Note that `Metadata::printAsOperand()` previously took an optional `bool` and `Module` operand. The former was cargo-culted from `Value::printAsOperand()` and wasn't doing anything useful. The latter didn't give consistent results (without the new `SlotTracker` mode). llvm-svn: 232275
2015-03-14 21:19:36 +01:00
WriteAsOperandInternal(OS, &MD, &TypePrinter, MST.getMachine(), M,
IR: Make Metadata::print() reliable and useful Replumb the `AsmWriter` so that `Metadata::print()` is generally useful. (Similarly change `Metadata::printAsOperand()`.) - `SlotTracker` now has a mode where all metadata will be correctly numbered when initializing a `Module`. Normally, `Metadata` only referenced from within `Function`s gets numbered when the `Function` is incorporated. - `Metadata::print()` and `Metadata::printAsOperand()` (and `Metadata::dump()`) now take an optional `Module` argument. When provided, `SlotTracker` is initialized with the new mode, and the numbering will be complete and consistent for all calls to `print()`. - `Value::print()` uses the new `SlotTracker` mode when printing intrinsics with `MDNode` operands, `MetadataAsValue` operands, or the bodies of functions. Thus, metadata numbering will be consistent between calls to `Metadata::print()` and `Value::print()`. - `Metadata::print()` (and `Metadata::dump()`) now print the full definition of `MDNode`s: !5 = !{!6, !"abc", !7} This matches behaviour for `Value::print()`, which includes the name of instructions. - Updated call sites in `Verifier` to call `print()` instead of `printAsOperand()`. All this, so that `Verifier` can print out useful failure messages that involve `Metadata` for PR22777. Note that `Metadata::printAsOperand()` previously took an optional `bool` and `Module` operand. The former was cargo-culted from `Value::printAsOperand()` and wasn't doing anything useful. The latter didn't give consistent results (without the new `SlotTracker` mode). llvm-svn: 232275
2015-03-14 21:19:36 +01:00
/* FromValue */ true);
auto *N = dyn_cast<MDNode>(&MD);
if (OnlyAsOperand || !N || isa<DIExpression>(MD))
IR: Split Metadata from Value Split `Metadata` away from the `Value` class hierarchy, as part of PR21532. Assembly and bitcode changes are in the wings, but this is the bulk of the change for the IR C++ API. I have a follow-up patch prepared for `clang`. If this breaks other sub-projects, I apologize in advance :(. Help me compile it on Darwin I'll try to fix it. FWIW, the errors should be easy to fix, so it may be simpler to just fix it yourself. This breaks the build for all metadata-related code that's out-of-tree. Rest assured the transition is mechanical and the compiler should catch almost all of the problems. Here's a quick guide for updating your code: - `Metadata` is the root of a class hierarchy with three main classes: `MDNode`, `MDString`, and `ValueAsMetadata`. It is distinct from the `Value` class hierarchy. It is typeless -- i.e., instances do *not* have a `Type`. - `MDNode`'s operands are all `Metadata *` (instead of `Value *`). - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively. If you're referring solely to resolved `MDNode`s -- post graph construction -- just use `MDNode*`. - `MDNode` (and the rest of `Metadata`) have only limited support for `replaceAllUsesWith()`. As long as an `MDNode` is pointing at a forward declaration -- the result of `MDNode::getTemporary()` -- it maintains a side map of its uses and can RAUW itself. Once the forward declarations are fully resolved RAUW support is dropped on the ground. This means that uniquing collisions on changing operands cause nodes to become "distinct". (This already happened fairly commonly, whenever an operand went to null.) If you're constructing complex (non self-reference) `MDNode` cycles, you need to call `MDNode::resolveCycles()` on each node (or on a top-level node that somehow references all of the nodes). Also, don't do that. Metadata cycles (and the RAUW machinery needed to construct them) are expensive. - An `MDNode` can only refer to a `Constant` through a bridge called `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`). As a side effect, accessing an operand of an `MDNode` that is known to be, e.g., `ConstantInt`, takes three steps: first, cast from `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`; third, cast down to `ConstantInt`. The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have metadata schema owners transition away from using `Constant`s when the type isn't important (and they don't care about referring to `GlobalValue`s). In the meantime, I've added transitional API to the `mdconst` namespace that matches semantics with the old code, in order to avoid adding the error-prone three-step equivalent to every call site. If your old code was: MDNode *N = foo(); bar(isa <ConstantInt>(N->getOperand(0))); baz(cast <ConstantInt>(N->getOperand(1))); bak(cast_or_null <ConstantInt>(N->getOperand(2))); bat(dyn_cast <ConstantInt>(N->getOperand(3))); bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4))); you can trivially match its semantics with: MDNode *N = foo(); bar(mdconst::hasa <ConstantInt>(N->getOperand(0))); baz(mdconst::extract <ConstantInt>(N->getOperand(1))); bak(mdconst::extract_or_null <ConstantInt>(N->getOperand(2))); bat(mdconst::dyn_extract <ConstantInt>(N->getOperand(3))); bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4))); and when you transition your metadata schema to `MDInt`: MDNode *N = foo(); bar(isa <MDInt>(N->getOperand(0))); baz(cast <MDInt>(N->getOperand(1))); bak(cast_or_null <MDInt>(N->getOperand(2))); bat(dyn_cast <MDInt>(N->getOperand(3))); bay(dyn_cast_or_null<MDInt>(N->getOperand(4))); - A `CallInst` -- specifically, intrinsic instructions -- can refer to metadata through a bridge called `MetadataAsValue`. This is a subclass of `Value` where `getType()->isMetadataTy()`. `MetadataAsValue` is the *only* class that can legally refer to a `LocalAsMetadata`, which is a bridged form of non-`Constant` values like `Argument` and `Instruction`. It can also refer to any other `Metadata` subclass. (I'll break all your testcases in a follow-up commit, when I propagate this change to assembly.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
return;
IR: Make Metadata::print() reliable and useful Replumb the `AsmWriter` so that `Metadata::print()` is generally useful. (Similarly change `Metadata::printAsOperand()`.) - `SlotTracker` now has a mode where all metadata will be correctly numbered when initializing a `Module`. Normally, `Metadata` only referenced from within `Function`s gets numbered when the `Function` is incorporated. - `Metadata::print()` and `Metadata::printAsOperand()` (and `Metadata::dump()`) now take an optional `Module` argument. When provided, `SlotTracker` is initialized with the new mode, and the numbering will be complete and consistent for all calls to `print()`. - `Value::print()` uses the new `SlotTracker` mode when printing intrinsics with `MDNode` operands, `MetadataAsValue` operands, or the bodies of functions. Thus, metadata numbering will be consistent between calls to `Metadata::print()` and `Value::print()`. - `Metadata::print()` (and `Metadata::dump()`) now print the full definition of `MDNode`s: !5 = !{!6, !"abc", !7} This matches behaviour for `Value::print()`, which includes the name of instructions. - Updated call sites in `Verifier` to call `print()` instead of `printAsOperand()`. All this, so that `Verifier` can print out useful failure messages that involve `Metadata` for PR22777. Note that `Metadata::printAsOperand()` previously took an optional `bool` and `Module` operand. The former was cargo-culted from `Value::printAsOperand()` and wasn't doing anything useful. The latter didn't give consistent results (without the new `SlotTracker` mode). llvm-svn: 232275
2015-03-14 21:19:36 +01:00
OS << " = ";
WriteMDNodeBodyInternal(OS, N, &TypePrinter, MST.getMachine(), M);
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
}
IR: Make Metadata::print() reliable and useful Replumb the `AsmWriter` so that `Metadata::print()` is generally useful. (Similarly change `Metadata::printAsOperand()`.) - `SlotTracker` now has a mode where all metadata will be correctly numbered when initializing a `Module`. Normally, `Metadata` only referenced from within `Function`s gets numbered when the `Function` is incorporated. - `Metadata::print()` and `Metadata::printAsOperand()` (and `Metadata::dump()`) now take an optional `Module` argument. When provided, `SlotTracker` is initialized with the new mode, and the numbering will be complete and consistent for all calls to `print()`. - `Value::print()` uses the new `SlotTracker` mode when printing intrinsics with `MDNode` operands, `MetadataAsValue` operands, or the bodies of functions. Thus, metadata numbering will be consistent between calls to `Metadata::print()` and `Value::print()`. - `Metadata::print()` (and `Metadata::dump()`) now print the full definition of `MDNode`s: !5 = !{!6, !"abc", !7} This matches behaviour for `Value::print()`, which includes the name of instructions. - Updated call sites in `Verifier` to call `print()` instead of `printAsOperand()`. All this, so that `Verifier` can print out useful failure messages that involve `Metadata` for PR22777. Note that `Metadata::printAsOperand()` previously took an optional `bool` and `Module` operand. The former was cargo-culted from `Value::printAsOperand()` and wasn't doing anything useful. The latter didn't give consistent results (without the new `SlotTracker` mode). llvm-svn: 232275
2015-03-14 21:19:36 +01:00
void Metadata::printAsOperand(raw_ostream &OS, const Module *M) const {
ModuleSlotTracker MST(M, isa<MDNode>(this));
printMetadataImpl(OS, *this, MST, M, /* OnlyAsOperand */ true);
}
void Metadata::printAsOperand(raw_ostream &OS, ModuleSlotTracker &MST,
const Module *M) const {
printMetadataImpl(OS, *this, MST, M, /* OnlyAsOperand */ true);
IR: Make Metadata::print() reliable and useful Replumb the `AsmWriter` so that `Metadata::print()` is generally useful. (Similarly change `Metadata::printAsOperand()`.) - `SlotTracker` now has a mode where all metadata will be correctly numbered when initializing a `Module`. Normally, `Metadata` only referenced from within `Function`s gets numbered when the `Function` is incorporated. - `Metadata::print()` and `Metadata::printAsOperand()` (and `Metadata::dump()`) now take an optional `Module` argument. When provided, `SlotTracker` is initialized with the new mode, and the numbering will be complete and consistent for all calls to `print()`. - `Value::print()` uses the new `SlotTracker` mode when printing intrinsics with `MDNode` operands, `MetadataAsValue` operands, or the bodies of functions. Thus, metadata numbering will be consistent between calls to `Metadata::print()` and `Value::print()`. - `Metadata::print()` (and `Metadata::dump()`) now print the full definition of `MDNode`s: !5 = !{!6, !"abc", !7} This matches behaviour for `Value::print()`, which includes the name of instructions. - Updated call sites in `Verifier` to call `print()` instead of `printAsOperand()`. All this, so that `Verifier` can print out useful failure messages that involve `Metadata` for PR22777. Note that `Metadata::printAsOperand()` previously took an optional `bool` and `Module` operand. The former was cargo-culted from `Value::printAsOperand()` and wasn't doing anything useful. The latter didn't give consistent results (without the new `SlotTracker` mode). llvm-svn: 232275
2015-03-14 21:19:36 +01:00
}
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
void Metadata::print(raw_ostream &OS, const Module *M,
bool /*IsForDebug*/) const {
ModuleSlotTracker MST(M, isa<MDNode>(this));
printMetadataImpl(OS, *this, MST, M, /* OnlyAsOperand */ 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.) llvm-svn: 223802
2014-12-09 19:38:53 +01:00
}
void Metadata::print(raw_ostream &OS, ModuleSlotTracker &MST,
const Module *M, bool /*IsForDebug*/) const {
printMetadataImpl(OS, *this, MST, M, /* OnlyAsOperand */ false);
}
#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
// Value::dump - allow easy printing of Values from the debugger.
LLVM_DUMP_METHOD
void Value::dump() const { print(dbgs(), /*IsForDebug=*/true); dbgs() << '\n'; }
// Type::dump - allow easy printing of Types from the debugger.
LLVM_DUMP_METHOD
void Type::dump() const { print(dbgs(), /*IsForDebug=*/true); dbgs() << '\n'; }
// Module::dump() - Allow printing of Modules from the debugger.
LLVM_DUMP_METHOD
void Module::dump() const {
print(dbgs(), nullptr,
/*ShouldPreserveUseListOrder=*/false, /*IsForDebug=*/true);
}
// Allow printing of Comdats from the debugger.
LLVM_DUMP_METHOD
void Comdat::dump() const { print(dbgs(), /*IsForDebug=*/true); }
// NamedMDNode::dump() - Allow printing of NamedMDNodes from the debugger.
LLVM_DUMP_METHOD
void NamedMDNode::dump() const { print(dbgs(), /*IsForDebug=*/true); }
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
LLVM_DUMP_METHOD
void Metadata::dump() const { dump(nullptr); }
LLVM_DUMP_METHOD
IR: Make Metadata::print() reliable and useful Replumb the `AsmWriter` so that `Metadata::print()` is generally useful. (Similarly change `Metadata::printAsOperand()`.) - `SlotTracker` now has a mode where all metadata will be correctly numbered when initializing a `Module`. Normally, `Metadata` only referenced from within `Function`s gets numbered when the `Function` is incorporated. - `Metadata::print()` and `Metadata::printAsOperand()` (and `Metadata::dump()`) now take an optional `Module` argument. When provided, `SlotTracker` is initialized with the new mode, and the numbering will be complete and consistent for all calls to `print()`. - `Value::print()` uses the new `SlotTracker` mode when printing intrinsics with `MDNode` operands, `MetadataAsValue` operands, or the bodies of functions. Thus, metadata numbering will be consistent between calls to `Metadata::print()` and `Value::print()`. - `Metadata::print()` (and `Metadata::dump()`) now print the full definition of `MDNode`s: !5 = !{!6, !"abc", !7} This matches behaviour for `Value::print()`, which includes the name of instructions. - Updated call sites in `Verifier` to call `print()` instead of `printAsOperand()`. All this, so that `Verifier` can print out useful failure messages that involve `Metadata` for PR22777. Note that `Metadata::printAsOperand()` previously took an optional `bool` and `Module` operand. The former was cargo-culted from `Value::printAsOperand()` and wasn't doing anything useful. The latter didn't give consistent results (without the new `SlotTracker` mode). llvm-svn: 232275
2015-03-14 21:19:36 +01:00
void Metadata::dump(const Module *M) const {
print(dbgs(), M, /*IsForDebug=*/true);
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
dbgs() << '\n';
}
#endif