1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-26 04:32:44 +01:00
llvm-mirror/utils/TableGen/InstrInfoEmitter.cpp

1080 lines
39 KiB
C++
Raw Normal View History

//===- InstrInfoEmitter.cpp - Generate a Instruction Set Desc. --*- C++ -*-===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This tablegen backend is responsible for emitting a description of the target
// instruction set for the code generator.
//
//===----------------------------------------------------------------------===//
#include "CodeGenDAGPatterns.h"
#include "CodeGenInstruction.h"
#include "CodeGenSchedule.h"
#include "CodeGenTarget.h"
#include "PredicateExpander.h"
#include "SequenceToOffsetTable.h"
#include "TableGenBackends.h"
#include "llvm/ADT/ArrayRef.h"
[M68k][TableGen](1/8) TableGen related changes - Add a new TableGen backend: CodeBeads - Add support to generate logical operand information For the first item, it is currently a workaround of M68k's (complex) instruction encoding. A typical architecture, especially CISC one like X86, normally uses `MCInstrDesc::TSFlags` to carry instruction encoding info. However, at the early days of M68k backend development, we found it difficult to fit every possible encoding into the 64-bit `MCInstrDesc::TSFlags`. Therefore CodeBeads was invented to provide an alternative, arbitrary length container for instruciton encoding info. However, in the long term we incline not to use a new TG backend for less common pattern like what we encountered in M68k. A bug has been created to host to discussion on migrating from CodeBeads to more concise solution: https://bugs.llvm.org/show_bug.cgi?id=48792 The second item was also served for similar purpose. It created utility functions that tell you the index of a `MachineOperand` in a `MachineInst` given a logical operand index. In normal cases a logical operand is the same as `MachineOperand`, but for operands using complex addressing mode a logical operand might be consisting of multiple `MachineOperand`. The TableGen-ed `getLogicalOperandIdx`, for instance, can give you the mapping between these two concepts. Nevertheless, we hope to remove this feature in the future if possible. Since it's not really useful for the targets supported by LLVM now either. Authors: myhsu, m4yers, glaubitz Differential Revision: https://reviews.llvm.org/D88385
2021-03-08 01:30:54 +01:00
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/StringExtras.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/TableGen/Error.h"
#include "llvm/TableGen/Record.h"
#include "llvm/TableGen/TableGenBackend.h"
#include <cassert>
#include <cstdint>
[M68k][TableGen](1/8) TableGen related changes - Add a new TableGen backend: CodeBeads - Add support to generate logical operand information For the first item, it is currently a workaround of M68k's (complex) instruction encoding. A typical architecture, especially CISC one like X86, normally uses `MCInstrDesc::TSFlags` to carry instruction encoding info. However, at the early days of M68k backend development, we found it difficult to fit every possible encoding into the 64-bit `MCInstrDesc::TSFlags`. Therefore CodeBeads was invented to provide an alternative, arbitrary length container for instruciton encoding info. However, in the long term we incline not to use a new TG backend for less common pattern like what we encountered in M68k. A bug has been created to host to discussion on migrating from CodeBeads to more concise solution: https://bugs.llvm.org/show_bug.cgi?id=48792 The second item was also served for similar purpose. It created utility functions that tell you the index of a `MachineOperand` in a `MachineInst` given a logical operand index. In normal cases a logical operand is the same as `MachineOperand`, but for operands using complex addressing mode a logical operand might be consisting of multiple `MachineOperand`. The TableGen-ed `getLogicalOperandIdx`, for instance, can give you the mapping between these two concepts. Nevertheless, we hope to remove this feature in the future if possible. Since it's not really useful for the targets supported by LLVM now either. Authors: myhsu, m4yers, glaubitz Differential Revision: https://reviews.llvm.org/D88385
2021-03-08 01:30:54 +01:00
#include <iterator>
#include <map>
#include <string>
#include <utility>
#include <vector>
using namespace llvm;
namespace {
class InstrInfoEmitter {
RecordKeeper &Records;
CodeGenDAGPatterns CDP;
const CodeGenSchedModels &SchedModels;
public:
InstrInfoEmitter(RecordKeeper &R):
Records(R), CDP(R), SchedModels(CDP.getTargetInfo().getSchedModels()) {}
// run - Output the instruction set description.
void run(raw_ostream &OS);
private:
void emitEnums(raw_ostream &OS);
typedef std::map<std::vector<std::string>, unsigned> OperandInfoMapTy;
/// The keys of this map are maps which have OpName enum values as their keys
/// and instruction operand indices as their values. The values of this map
/// are lists of instruction names.
typedef std::map<std::map<unsigned, unsigned>,
std::vector<std::string>> OpNameMapTy;
typedef std::map<std::string, unsigned>::iterator StrUintMapIter;
/// Generate member functions in the target-specific GenInstrInfo class.
///
/// This method is used to custom expand TIIPredicate definitions.
/// See file llvm/Target/TargetInstPredicates.td for a description of what is
/// a TIIPredicate and how to use it.
[tblgen][PredicateExpander] Add the ability to describe more complex constraints on instruction operands. Before this patch, class PredicateExpander only knew how to expand simple predicates that performed checks on instruction operands. In particular, the new scheduling predicate syntax was not rich enough to express checks like this one: Foo(MI->getOperand(0).getImm()) == ExpectedVal; Here, the immediate operand value at index zero is passed in input to function Foo, and ExpectedVal is compared against the value returned by function Foo. While this predicate pattern doesn't show up in any X86 model, it shows up in other upstream targets. So, being able to support those predicates is fundamental if we want to be able to modernize all the scheduling models upstream. With this patch, we allow users to specify if a register/immediate operand value needs to be passed in input to a function as part of the predicate check. Now, register/immediate operand checks all derive from base class CheckOperandBase. This patch also changes where TIIPredicate definitions are expanded by the instructon info emitter. Before, definitions were expanded in class XXXGenInstrInfo (where XXX is a target name). With the introduction of this new syntax, we may want to have TIIPredicates expanded directly in XXXInstrInfo. That is because functions used by the new operand predicates may only exist in the derived class (i.e. XXXInstrInfo). This patch is a non functional change for the existing scheduling models. In future, we will be able to use this richer syntax to better describe complex scheduling predicates, and expose them to llvm-mca. Differential Revision: https://reviews.llvm.org/D53880 llvm-svn: 345714
2018-10-31 13:28:05 +01:00
void emitTIIHelperMethods(raw_ostream &OS, StringRef TargetName,
bool ExpandDefinition = true);
/// Expand TIIPredicate definitions to functions that accept a const MCInst
/// reference.
void emitMCIIHelperMethods(raw_ostream &OS, StringRef TargetName);
void emitRecord(const CodeGenInstruction &Inst, unsigned Num,
Record *InstrInfo,
std::map<std::vector<Record*>, unsigned> &EL,
const OperandInfoMapTy &OpInfo,
raw_ostream &OS);
void emitOperandTypeMappings(
raw_ostream &OS, const CodeGenTarget &Target,
ArrayRef<const CodeGenInstruction *> NumberedInstructions);
void initOperandMapData(
ArrayRef<const CodeGenInstruction *> NumberedInstructions,
StringRef Namespace,
std::map<std::string, unsigned> &Operands,
OpNameMapTy &OperandMap);
void emitOperandNameMappings(raw_ostream &OS, const CodeGenTarget &Target,
ArrayRef<const CodeGenInstruction*> NumberedInstructions);
[M68k][TableGen](1/8) TableGen related changes - Add a new TableGen backend: CodeBeads - Add support to generate logical operand information For the first item, it is currently a workaround of M68k's (complex) instruction encoding. A typical architecture, especially CISC one like X86, normally uses `MCInstrDesc::TSFlags` to carry instruction encoding info. However, at the early days of M68k backend development, we found it difficult to fit every possible encoding into the 64-bit `MCInstrDesc::TSFlags`. Therefore CodeBeads was invented to provide an alternative, arbitrary length container for instruciton encoding info. However, in the long term we incline not to use a new TG backend for less common pattern like what we encountered in M68k. A bug has been created to host to discussion on migrating from CodeBeads to more concise solution: https://bugs.llvm.org/show_bug.cgi?id=48792 The second item was also served for similar purpose. It created utility functions that tell you the index of a `MachineOperand` in a `MachineInst` given a logical operand index. In normal cases a logical operand is the same as `MachineOperand`, but for operands using complex addressing mode a logical operand might be consisting of multiple `MachineOperand`. The TableGen-ed `getLogicalOperandIdx`, for instance, can give you the mapping between these two concepts. Nevertheless, we hope to remove this feature in the future if possible. Since it's not really useful for the targets supported by LLVM now either. Authors: myhsu, m4yers, glaubitz Differential Revision: https://reviews.llvm.org/D88385
2021-03-08 01:30:54 +01:00
void emitLogicalOperandSizeMappings(
raw_ostream &OS, StringRef Namespace,
ArrayRef<const CodeGenInstruction *> NumberedInstructions);
void emitLogicalOperandTypeMappings(
raw_ostream &OS, StringRef Namespace,
ArrayRef<const CodeGenInstruction *> NumberedInstructions);
// Operand information.
void EmitOperandInfo(raw_ostream &OS, OperandInfoMapTy &OperandInfoIDs);
std::vector<std::string> GetOperandInfo(const CodeGenInstruction &Inst);
};
} // end anonymous namespace
2008-01-06 02:21:51 +01:00
static void PrintDefList(const std::vector<Record*> &Uses,
unsigned Num, raw_ostream &OS) {
OS << "static const MCPhysReg ImplicitList" << Num << "[] = { ";
for (Record *U : Uses)
OS << getQualifiedName(U) << ", ";
OS << "0 };\n";
}
//===----------------------------------------------------------------------===//
// Operand Info Emission.
//===----------------------------------------------------------------------===//
std::vector<std::string>
InstrInfoEmitter::GetOperandInfo(const CodeGenInstruction &Inst) {
std::vector<std::string> Result;
for (auto &Op : Inst.Operands) {
// Handle aggregate operands and normal operands the same way by expanding
// either case into a list of operands for this op.
std::vector<CGIOperandList::OperandInfo> OperandList;
// This might be a multiple operand thing. Targets like X86 have
// registers in their multi-operand operands. It may also be an anonymous
// operand, which has a single operand, but no declared class for the
// operand.
DagInit *MIOI = Op.MIOperandInfo;
if (!MIOI || MIOI->getNumArgs() == 0) {
// Single, anonymous, operand.
OperandList.push_back(Op);
} else {
for (unsigned j = 0, e = Op.MINumOperands; j != e; ++j) {
OperandList.push_back(Op);
auto *OpR = cast<DefInit>(MIOI->getArg(j))->getDef();
OperandList.back().Rec = OpR;
}
}
for (unsigned j = 0, e = OperandList.size(); j != e; ++j) {
Record *OpR = OperandList[j].Rec;
std::string Res;
if (OpR->isSubClassOf("RegisterOperand"))
OpR = OpR->getValueAsDef("RegClass");
if (OpR->isSubClassOf("RegisterClass"))
Res += getQualifiedName(OpR) + "RegClassID, ";
else if (OpR->isSubClassOf("PointerLikeRegClass"))
Res += utostr(OpR->getValueAsInt("RegClassKind")) + ", ";
else
// -1 means the operand does not have a fixed register class.
Res += "-1, ";
// Fill in applicable flags.
Res += "0";
// Ptr value whose register class is resolved via callback.
if (OpR->isSubClassOf("PointerLikeRegClass"))
Res += "|(1<<MCOI::LookupPtrRegClass)";
// Predicate operands. Check to see if the original unexpanded operand
// was of type PredicateOp.
if (Op.Rec->isSubClassOf("PredicateOp"))
Res += "|(1<<MCOI::Predicate)";
// Optional def operands. Check to see if the original unexpanded operand
// was of type OptionalDefOperand.
if (Op.Rec->isSubClassOf("OptionalDefOperand"))
Res += "|(1<<MCOI::OptionalDef)";
// Branch target operands. Check to see if the original unexpanded
// operand was of type BranchTargetOperand.
if (Op.Rec->isSubClassOf("BranchTargetOperand"))
Res += "|(1<<MCOI::BranchTarget)";
// Fill in operand type.
Res += ", ";
assert(!Op.OperandType.empty() && "Invalid operand type.");
Res += Op.OperandType;
// Fill in constraint info.
Res += ", ";
const CGIOperandList::ConstraintInfo &Constraint =
Op.Constraints[j];
if (Constraint.isNone())
Res += "0";
else if (Constraint.isEarlyClobber())
Res += "MCOI_EARLY_CLOBBER";
else {
assert(Constraint.isTied());
Res += "MCOI_TIED_TO(" + utostr(Constraint.getTiedOperand()) + ")";
}
Result.push_back(Res);
}
}
return Result;
}
void InstrInfoEmitter::EmitOperandInfo(raw_ostream &OS,
OperandInfoMapTy &OperandInfoIDs) {
// ID #0 is for no operand info.
unsigned OperandListNum = 0;
OperandInfoIDs[std::vector<std::string>()] = ++OperandListNum;
OS << "\n";
const CodeGenTarget &Target = CDP.getTargetInfo();
for (const CodeGenInstruction *Inst : Target.getInstructionsByEnumValue()) {
std::vector<std::string> OperandInfo = GetOperandInfo(*Inst);
unsigned &N = OperandInfoIDs[OperandInfo];
if (N != 0) continue;
N = ++OperandListNum;
OS << "static const MCOperandInfo OperandInfo" << N << "[] = { ";
for (const std::string &Info : OperandInfo)
OS << "{ " << Info << " }, ";
OS << "};\n";
}
}
/// Initialize data structures for generating operand name mappings.
///
/// \param Operands [out] A map used to generate the OpName enum with operand
/// names as its keys and operand enum values as its values.
/// \param OperandMap [out] A map for representing the operand name mappings for
/// each instructions. This is used to generate the OperandMap table as
/// well as the getNamedOperandIdx() function.
void InstrInfoEmitter::initOperandMapData(
ArrayRef<const CodeGenInstruction *> NumberedInstructions,
StringRef Namespace,
std::map<std::string, unsigned> &Operands,
OpNameMapTy &OperandMap) {
unsigned NumOperands = 0;
for (const CodeGenInstruction *Inst : NumberedInstructions) {
if (!Inst->TheDef->getValueAsBit("UseNamedOperandTable"))
continue;
std::map<unsigned, unsigned> OpList;
for (const auto &Info : Inst->Operands) {
StrUintMapIter I = Operands.find(Info.Name);
if (I == Operands.end()) {
I = Operands.insert(Operands.begin(),
std::pair<std::string, unsigned>(Info.Name, NumOperands++));
}
OpList[I->second] = Info.MIOperandNo;
}
OperandMap[OpList].push_back(Namespace.str() + "::" +
Inst->TheDef->getName().str());
}
}
/// Generate a table and function for looking up the indices of operands by
/// name.
///
/// This code generates:
/// - An enum in the llvm::TargetNamespace::OpName namespace, with one entry
/// for each operand name.
/// - A 2-dimensional table called OperandMap for mapping OpName enum values to
/// operand indices.
/// - A function called getNamedOperandIdx(uint16_t Opcode, uint16_t NamedIdx)
/// for looking up the operand index for an instruction, given a value from
/// OpName enum
void InstrInfoEmitter::emitOperandNameMappings(raw_ostream &OS,
const CodeGenTarget &Target,
ArrayRef<const CodeGenInstruction*> NumberedInstructions) {
StringRef Namespace = Target.getInstNamespace();
std::string OpNameNS = "OpName";
// Map of operand names to their enumeration value. This will be used to
// generate the OpName enum.
std::map<std::string, unsigned> Operands;
OpNameMapTy OperandMap;
initOperandMapData(NumberedInstructions, Namespace, Operands, OperandMap);
OS << "#ifdef GET_INSTRINFO_OPERAND_ENUM\n";
OS << "#undef GET_INSTRINFO_OPERAND_ENUM\n";
OS << "namespace llvm {\n";
OS << "namespace " << Namespace << " {\n";
OS << "namespace " << OpNameNS << " {\n";
OS << "enum {\n";
for (const auto &Op : Operands)
OS << " " << Op.first << " = " << Op.second << ",\n";
OS << " OPERAND_LAST";
OS << "\n};\n";
OS << "} // end namespace OpName\n";
OS << "} // end namespace " << Namespace << "\n";
OS << "} // end namespace llvm\n";
OS << "#endif //GET_INSTRINFO_OPERAND_ENUM\n\n";
OS << "#ifdef GET_INSTRINFO_NAMED_OPS\n";
OS << "#undef GET_INSTRINFO_NAMED_OPS\n";
OS << "namespace llvm {\n";
OS << "namespace " << Namespace << " {\n";
OS << "LLVM_READONLY\n";
OS << "int16_t getNamedOperandIdx(uint16_t Opcode, uint16_t NamedIdx) {\n";
if (!Operands.empty()) {
OS << " static const int16_t OperandMap [][" << Operands.size()
<< "] = {\n";
for (const auto &Entry : OperandMap) {
const std::map<unsigned, unsigned> &OpList = Entry.first;
OS << "{";
// Emit a row of the OperandMap table
for (unsigned i = 0, e = Operands.size(); i != e; ++i)
OS << (OpList.count(i) == 0 ? -1 : (int)OpList.find(i)->second) << ", ";
OS << "},\n";
}
OS << "};\n";
OS << " switch(Opcode) {\n";
unsigned TableIndex = 0;
for (const auto &Entry : OperandMap) {
for (const std::string &Name : Entry.second)
OS << " case " << Name << ":\n";
OS << " return OperandMap[" << TableIndex++ << "][NamedIdx];\n";
}
OS << " default: return -1;\n";
OS << " }\n";
} else {
// There are no operands, so no need to emit anything
OS << " return -1;\n";
}
OS << "}\n";
OS << "} // end namespace " << Namespace << "\n";
OS << "} // end namespace llvm\n";
OS << "#endif //GET_INSTRINFO_NAMED_OPS\n\n";
}
/// Generate an enum for all the operand types for this target, under the
/// llvm::TargetNamespace::OpTypes namespace.
/// Operand types are all definitions derived of the Operand Target.td class.
void InstrInfoEmitter::emitOperandTypeMappings(
raw_ostream &OS, const CodeGenTarget &Target,
ArrayRef<const CodeGenInstruction *> NumberedInstructions) {
StringRef Namespace = Target.getInstNamespace();
std::vector<Record *> Operands = Records.getAllDerivedDefinitions("Operand");
std::vector<Record *> RegisterOperands =
Records.getAllDerivedDefinitions("RegisterOperand");
std::vector<Record *> RegisterClasses =
Records.getAllDerivedDefinitions("RegisterClass");
OS << "#ifdef GET_INSTRINFO_OPERAND_TYPES_ENUM\n";
OS << "#undef GET_INSTRINFO_OPERAND_TYPES_ENUM\n";
OS << "namespace llvm {\n";
OS << "namespace " << Namespace << " {\n";
OS << "namespace OpTypes {\n";
OS << "enum OperandType {\n";
unsigned EnumVal = 0;
for (const std::vector<Record *> *RecordsToAdd :
{&Operands, &RegisterOperands, &RegisterClasses}) {
for (const Record *Op : *RecordsToAdd) {
if (!Op->isAnonymous())
OS << " " << Op->getName() << " = " << EnumVal << ",\n";
++EnumVal;
}
}
OS << " OPERAND_TYPE_LIST_END" << "\n};\n";
OS << "} // end namespace OpTypes\n";
OS << "} // end namespace " << Namespace << "\n";
OS << "} // end namespace llvm\n";
OS << "#endif // GET_INSTRINFO_OPERAND_TYPES_ENUM\n\n";
OS << "#ifdef GET_INSTRINFO_OPERAND_TYPE\n";
OS << "#undef GET_INSTRINFO_OPERAND_TYPE\n";
OS << "namespace llvm {\n";
OS << "namespace " << Namespace << " {\n";
OS << "LLVM_READONLY\n";
OS << "static int getOperandType(uint16_t Opcode, uint16_t OpIdx) {\n";
// TODO: Factor out duplicate operand lists to compress the tables.
if (!NumberedInstructions.empty()) {
std::vector<int> OperandOffsets;
std::vector<Record *> OperandRecords;
int CurrentOffset = 0;
for (const CodeGenInstruction *Inst : NumberedInstructions) {
OperandOffsets.push_back(CurrentOffset);
for (const auto &Op : Inst->Operands) {
const DagInit *MIOI = Op.MIOperandInfo;
if (!MIOI || MIOI->getNumArgs() == 0) {
// Single, anonymous, operand.
OperandRecords.push_back(Op.Rec);
++CurrentOffset;
} else {
for (Init *Arg : MIOI->getArgs()) {
OperandRecords.push_back(cast<DefInit>(Arg)->getDef());
++CurrentOffset;
}
}
}
}
// Emit the table of offsets (indexes) into the operand type table.
// Size the unsigned integer offset to save space.
assert(OperandRecords.size() <= UINT32_MAX &&
"Too many operands for offset table");
OS << ((OperandRecords.size() <= UINT16_MAX) ? " const uint16_t"
: " const uint32_t");
OS << " Offsets[] = {\n";
for (int I = 0, E = OperandOffsets.size(); I != E; ++I)
OS << " " << OperandOffsets[I] << ",\n";
OS << " };\n";
// Add an entry for the end so that we don't need to special case it below.
OperandOffsets.push_back(OperandRecords.size());
// Emit the actual operand types in a flat table.
// Size the signed integer operand type to save space.
assert(EnumVal <= INT16_MAX &&
"Too many operand types for operand types table");
OS << ((EnumVal <= INT8_MAX) ? " const int8_t" : " const int16_t");
OS << " OpcodeOperandTypes[] = {\n ";
for (int I = 0, E = OperandRecords.size(), CurOffset = 1; I != E; ++I) {
// We print each Opcode's operands in its own row.
if (I == OperandOffsets[CurOffset]) {
OS << "\n ";
// If there are empty rows, mark them with an empty comment.
while (OperandOffsets[++CurOffset] == I)
OS << "/**/\n ";
}
Record *OpR = OperandRecords[I];
if ((OpR->isSubClassOf("Operand") ||
OpR->isSubClassOf("RegisterOperand") ||
OpR->isSubClassOf("RegisterClass")) &&
!OpR->isAnonymous())
OS << "OpTypes::" << OpR->getName();
else
OS << -1;
OS << ", ";
}
OS << "\n };\n";
OS << " return OpcodeOperandTypes[Offsets[Opcode] + OpIdx];\n";
} else {
OS << " llvm_unreachable(\"No instructions defined\");\n";
}
OS << "}\n";
OS << "} // end namespace " << Namespace << "\n";
OS << "} // end namespace llvm\n";
OS << "#endif // GET_INSTRINFO_OPERAND_TYPE\n\n";
}
[M68k][TableGen](1/8) TableGen related changes - Add a new TableGen backend: CodeBeads - Add support to generate logical operand information For the first item, it is currently a workaround of M68k's (complex) instruction encoding. A typical architecture, especially CISC one like X86, normally uses `MCInstrDesc::TSFlags` to carry instruction encoding info. However, at the early days of M68k backend development, we found it difficult to fit every possible encoding into the 64-bit `MCInstrDesc::TSFlags`. Therefore CodeBeads was invented to provide an alternative, arbitrary length container for instruciton encoding info. However, in the long term we incline not to use a new TG backend for less common pattern like what we encountered in M68k. A bug has been created to host to discussion on migrating from CodeBeads to more concise solution: https://bugs.llvm.org/show_bug.cgi?id=48792 The second item was also served for similar purpose. It created utility functions that tell you the index of a `MachineOperand` in a `MachineInst` given a logical operand index. In normal cases a logical operand is the same as `MachineOperand`, but for operands using complex addressing mode a logical operand might be consisting of multiple `MachineOperand`. The TableGen-ed `getLogicalOperandIdx`, for instance, can give you the mapping between these two concepts. Nevertheless, we hope to remove this feature in the future if possible. Since it's not really useful for the targets supported by LLVM now either. Authors: myhsu, m4yers, glaubitz Differential Revision: https://reviews.llvm.org/D88385
2021-03-08 01:30:54 +01:00
void InstrInfoEmitter::emitLogicalOperandSizeMappings(
raw_ostream &OS, StringRef Namespace,
ArrayRef<const CodeGenInstruction *> NumberedInstructions) {
std::map<std::vector<unsigned>, unsigned> LogicalOpSizeMap;
std::map<unsigned, std::vector<std::string>> InstMap;
size_t LogicalOpListSize = 0U;
std::vector<unsigned> LogicalOpList;
for (const auto *Inst : NumberedInstructions) {
if (!Inst->TheDef->getValueAsBit("UseLogicalOperandMappings"))
continue;
LogicalOpList.clear();
llvm::transform(Inst->Operands, std::back_inserter(LogicalOpList),
[](const CGIOperandList::OperandInfo &Op) -> unsigned {
auto *MIOI = Op.MIOperandInfo;
if (!MIOI || MIOI->getNumArgs() == 0)
return 1;
return MIOI->getNumArgs();
});
LogicalOpListSize = std::max(LogicalOpList.size(), LogicalOpListSize);
auto I =
LogicalOpSizeMap.insert({LogicalOpList, LogicalOpSizeMap.size()}).first;
InstMap[I->second].push_back(
(Namespace + "::" + Inst->TheDef->getName()).str());
}
OS << "#ifdef GET_INSTRINFO_LOGICAL_OPERAND_SIZE_MAP\n";
OS << "#undef GET_INSTRINFO_LOGICAL_OPERAND_SIZE_MAP\n";
OS << "namespace llvm {\n";
OS << "namespace " << Namespace << " {\n";
OS << "LLVM_READONLY static unsigned\n";
OS << "getLogicalOperandSize(uint16_t Opcode, uint16_t LogicalOpIdx) {\n";
if (!InstMap.empty()) {
std::vector<const std::vector<unsigned> *> LogicalOpSizeList(
LogicalOpSizeMap.size());
for (auto &P : LogicalOpSizeMap) {
LogicalOpSizeList[P.second] = &P.first;
}
OS << " static const unsigned SizeMap[][" << LogicalOpListSize
<< "] = {\n";
for (int r = 0, rs = LogicalOpSizeList.size(); r < rs; ++r) {
const auto &Row = *LogicalOpSizeList[r];
OS << " {";
int i;
for (i = 0; i < static_cast<int>(Row.size()); ++i) {
OS << Row[i] << ", ";
}
for (; i < static_cast<int>(LogicalOpListSize); ++i) {
OS << "0, ";
}
OS << "}, ";
OS << "\n";
}
OS << " };\n";
OS << " switch (Opcode) {\n";
OS << " default: return LogicalOpIdx;\n";
for (auto &P : InstMap) {
auto OpMapIdx = P.first;
const auto &Insts = P.second;
for (const auto &Inst : Insts) {
OS << " case " << Inst << ":\n";
}
OS << " return SizeMap[" << OpMapIdx << "][LogicalOpIdx];\n";
}
OS << " }\n";
} else {
OS << " return LogicalOpIdx;\n";
}
OS << "}\n";
OS << "LLVM_READONLY static inline unsigned\n";
OS << "getLogicalOperandIdx(uint16_t Opcode, uint16_t LogicalOpIdx) {\n";
OS << " auto S = 0U;\n";
OS << " for (auto i = 0U; i < LogicalOpIdx; ++i)\n";
OS << " S += getLogicalOperandSize(Opcode, i);\n";
OS << " return S;\n";
OS << "}\n";
OS << "} // end namespace " << Namespace << "\n";
OS << "} // end namespace llvm\n";
OS << "#endif // GET_INSTRINFO_LOGICAL_OPERAND_SIZE_MAP\n\n";
}
void InstrInfoEmitter::emitLogicalOperandTypeMappings(
raw_ostream &OS, StringRef Namespace,
ArrayRef<const CodeGenInstruction *> NumberedInstructions) {
std::map<std::vector<std::string>, unsigned> LogicalOpTypeMap;
std::map<unsigned, std::vector<std::string>> InstMap;
size_t OpTypeListSize = 0U;
std::vector<std::string> LogicalOpTypeList;
for (const auto *Inst : NumberedInstructions) {
if (!Inst->TheDef->getValueAsBit("UseLogicalOperandMappings"))
continue;
LogicalOpTypeList.clear();
for (const auto &Op : Inst->Operands) {
auto *OpR = Op.Rec;
if ((OpR->isSubClassOf("Operand") ||
OpR->isSubClassOf("RegisterOperand") ||
OpR->isSubClassOf("RegisterClass")) &&
!OpR->isAnonymous()) {
LogicalOpTypeList.push_back(
(Namespace + "::OpTypes::" + Op.Rec->getName()).str());
} else {
LogicalOpTypeList.push_back("-1");
}
}
OpTypeListSize = std::max(LogicalOpTypeList.size(), OpTypeListSize);
auto I =
LogicalOpTypeMap.insert({LogicalOpTypeList, LogicalOpTypeMap.size()})
.first;
InstMap[I->second].push_back(
(Namespace + "::" + Inst->TheDef->getName()).str());
}
OS << "#ifdef GET_INSTRINFO_LOGICAL_OPERAND_TYPE_MAP\n";
OS << "#undef GET_INSTRINFO_LOGICAL_OPERAND_TYPE_MAP\n";
OS << "namespace llvm {\n";
OS << "namespace " << Namespace << " {\n";
OS << "LLVM_READONLY static int\n";
OS << "getLogicalOperandType(uint16_t Opcode, uint16_t LogicalOpIdx) {\n";
if (!InstMap.empty()) {
std::vector<const std::vector<std::string> *> LogicalOpTypeList(
LogicalOpTypeMap.size());
for (auto &P : LogicalOpTypeMap) {
LogicalOpTypeList[P.second] = &P.first;
}
OS << " static const int TypeMap[][" << OpTypeListSize << "] = {\n";
for (int r = 0, rs = LogicalOpTypeList.size(); r < rs; ++r) {
const auto &Row = *LogicalOpTypeList[r];
OS << " {";
int i, s = Row.size();
for (i = 0; i < s; ++i) {
if (i > 0)
OS << ", ";
OS << Row[i];
}
for (; i < static_cast<int>(OpTypeListSize); ++i) {
if (i > 0)
OS << ", ";
OS << "-1";
}
OS << "}";
if (r != rs - 1)
OS << ",";
OS << "\n";
}
OS << " };\n";
OS << " switch (Opcode) {\n";
OS << " default: return -1;\n";
for (auto &P : InstMap) {
auto OpMapIdx = P.first;
const auto &Insts = P.second;
for (const auto &Inst : Insts) {
OS << " case " << Inst << ":\n";
}
OS << " return TypeMap[" << OpMapIdx << "][LogicalOpIdx];\n";
}
OS << " }\n";
} else {
OS << " return -1;\n";
}
OS << "}\n";
OS << "} // end namespace " << Namespace << "\n";
OS << "} // end namespace llvm\n";
OS << "#endif // GET_INSTRINFO_LOGICAL_OPERAND_TYPE_MAP\n\n";
}
void InstrInfoEmitter::emitMCIIHelperMethods(raw_ostream &OS,
StringRef TargetName) {
RecVec TIIPredicates = Records.getAllDerivedDefinitions("TIIPredicate");
if (TIIPredicates.empty())
return;
OS << "#ifdef GET_INSTRINFO_MC_HELPER_DECLS\n";
OS << "#undef GET_INSTRINFO_MC_HELPER_DECLS\n\n";
OS << "namespace llvm {\n";
OS << "class MCInst;\n\n";
OS << "namespace " << TargetName << "_MC {\n\n";
for (const Record *Rec : TIIPredicates) {
OS << "bool " << Rec->getValueAsString("FunctionName")
<< "(const MCInst &MI);\n";
}
OS << "\n} // end namespace " << TargetName << "_MC\n";
OS << "} // end namespace llvm\n\n";
OS << "#endif // GET_INSTRINFO_MC_HELPER_DECLS\n\n";
OS << "#ifdef GET_INSTRINFO_MC_HELPERS\n";
OS << "#undef GET_INSTRINFO_MC_HELPERS\n\n";
OS << "namespace llvm {\n";
OS << "namespace " << TargetName << "_MC {\n\n";
PredicateExpander PE(TargetName);
PE.setExpandForMC(true);
for (const Record *Rec : TIIPredicates) {
OS << "bool " << Rec->getValueAsString("FunctionName");
OS << "(const MCInst &MI) {\n";
OS.indent(PE.getIndentLevel() * 2);
PE.expandStatement(OS, Rec->getValueAsDef("Body"));
OS << "\n}\n\n";
}
OS << "} // end namespace " << TargetName << "_MC\n";
OS << "} // end namespace llvm\n\n";
OS << "#endif // GET_GENISTRINFO_MC_HELPERS\n";
}
void InstrInfoEmitter::emitTIIHelperMethods(raw_ostream &OS,
[tblgen][PredicateExpander] Add the ability to describe more complex constraints on instruction operands. Before this patch, class PredicateExpander only knew how to expand simple predicates that performed checks on instruction operands. In particular, the new scheduling predicate syntax was not rich enough to express checks like this one: Foo(MI->getOperand(0).getImm()) == ExpectedVal; Here, the immediate operand value at index zero is passed in input to function Foo, and ExpectedVal is compared against the value returned by function Foo. While this predicate pattern doesn't show up in any X86 model, it shows up in other upstream targets. So, being able to support those predicates is fundamental if we want to be able to modernize all the scheduling models upstream. With this patch, we allow users to specify if a register/immediate operand value needs to be passed in input to a function as part of the predicate check. Now, register/immediate operand checks all derive from base class CheckOperandBase. This patch also changes where TIIPredicate definitions are expanded by the instructon info emitter. Before, definitions were expanded in class XXXGenInstrInfo (where XXX is a target name). With the introduction of this new syntax, we may want to have TIIPredicates expanded directly in XXXInstrInfo. That is because functions used by the new operand predicates may only exist in the derived class (i.e. XXXInstrInfo). This patch is a non functional change for the existing scheduling models. In future, we will be able to use this richer syntax to better describe complex scheduling predicates, and expose them to llvm-mca. Differential Revision: https://reviews.llvm.org/D53880 llvm-svn: 345714
2018-10-31 13:28:05 +01:00
StringRef TargetName,
bool ExpandDefinition) {
RecVec TIIPredicates = Records.getAllDerivedDefinitions("TIIPredicate");
if (TIIPredicates.empty())
return;
PredicateExpander PE(TargetName);
PE.setExpandForMC(false);
for (const Record *Rec : TIIPredicates) {
OS << (ExpandDefinition ? "" : "static ") << "bool ";
[tblgen][PredicateExpander] Add the ability to describe more complex constraints on instruction operands. Before this patch, class PredicateExpander only knew how to expand simple predicates that performed checks on instruction operands. In particular, the new scheduling predicate syntax was not rich enough to express checks like this one: Foo(MI->getOperand(0).getImm()) == ExpectedVal; Here, the immediate operand value at index zero is passed in input to function Foo, and ExpectedVal is compared against the value returned by function Foo. While this predicate pattern doesn't show up in any X86 model, it shows up in other upstream targets. So, being able to support those predicates is fundamental if we want to be able to modernize all the scheduling models upstream. With this patch, we allow users to specify if a register/immediate operand value needs to be passed in input to a function as part of the predicate check. Now, register/immediate operand checks all derive from base class CheckOperandBase. This patch also changes where TIIPredicate definitions are expanded by the instructon info emitter. Before, definitions were expanded in class XXXGenInstrInfo (where XXX is a target name). With the introduction of this new syntax, we may want to have TIIPredicates expanded directly in XXXInstrInfo. That is because functions used by the new operand predicates may only exist in the derived class (i.e. XXXInstrInfo). This patch is a non functional change for the existing scheduling models. In future, we will be able to use this richer syntax to better describe complex scheduling predicates, and expose them to llvm-mca. Differential Revision: https://reviews.llvm.org/D53880 llvm-svn: 345714
2018-10-31 13:28:05 +01:00
if (ExpandDefinition)
OS << TargetName << "InstrInfo::";
OS << Rec->getValueAsString("FunctionName");
OS << "(const MachineInstr &MI)";
if (!ExpandDefinition) {
OS << ";\n";
continue;
}
OS << " {\n";
OS.indent(PE.getIndentLevel() * 2);
PE.expandStatement(OS, Rec->getValueAsDef("Body"));
OS << "\n}\n\n";
}
}
//===----------------------------------------------------------------------===//
// Main Output.
//===----------------------------------------------------------------------===//
// run - Emit the main instruction description records for the target...
void InstrInfoEmitter::run(raw_ostream &OS) {
emitSourceFileHeader("Target Instruction Enum Values and Descriptors", OS);
emitEnums(OS);
OS << "#ifdef GET_INSTRINFO_MC_DESC\n";
OS << "#undef GET_INSTRINFO_MC_DESC\n";
OS << "namespace llvm {\n\n";
CodeGenTarget &Target = CDP.getTargetInfo();
const std::string &TargetName = std::string(Target.getName());
Record *InstrInfo = Target.getInstructionSet();
// Keep track of all of the def lists we have emitted already.
std::map<std::vector<Record*>, unsigned> EmittedLists;
unsigned ListNumber = 0;
// Emit all of the instruction's implicit uses and defs.
Records.startTimer("Emit uses/defs");
for (const CodeGenInstruction *II : Target.getInstructionsByEnumValue()) {
Record *Inst = II->TheDef;
std::vector<Record*> Uses = Inst->getValueAsListOfDefs("Uses");
if (!Uses.empty()) {
unsigned &IL = EmittedLists[Uses];
2008-01-06 02:21:51 +01:00
if (!IL) PrintDefList(Uses, IL = ++ListNumber, OS);
}
std::vector<Record*> Defs = Inst->getValueAsListOfDefs("Defs");
if (!Defs.empty()) {
unsigned &IL = EmittedLists[Defs];
2008-01-06 02:21:51 +01:00
if (!IL) PrintDefList(Defs, IL = ++ListNumber, OS);
}
}
OperandInfoMapTy OperandInfoIDs;
// Emit all of the operand info records.
Records.startTimer("Emit operand info");
EmitOperandInfo(OS, OperandInfoIDs);
// Emit all of the MCInstrDesc records in their ENUM ordering.
//
Records.startTimer("Emit InstrDesc records");
OS << "\nextern const MCInstrDesc " << TargetName << "Insts[] = {\n";
ArrayRef<const CodeGenInstruction*> NumberedInstructions =
Target.getInstructionsByEnumValue();
SequenceToOffsetTable<std::string> InstrNames;
unsigned Num = 0;
for (const CodeGenInstruction *Inst : NumberedInstructions) {
// Keep a list of the instruction names.
InstrNames.add(std::string(Inst->TheDef->getName()));
// Emit the record into the table.
emitRecord(*Inst, Num++, InstrInfo, EmittedLists, OperandInfoIDs, OS);
}
OS << "};\n\n";
// Emit the array of instruction names.
Records.startTimer("Emit instruction names");
InstrNames.layout();
InstrNames.emitStringLiteralDef(OS, Twine("extern const char ") + TargetName +
"InstrNameData[]");
OS << "extern const unsigned " << TargetName <<"InstrNameIndices[] = {";
Num = 0;
for (const CodeGenInstruction *Inst : NumberedInstructions) {
// Newline every eight entries.
if (Num % 8 == 0)
OS << "\n ";
OS << InstrNames.get(std::string(Inst->TheDef->getName())) << "U, ";
++Num;
}
OS << "\n};\n\n";
bool HasDeprecationFeatures =
llvm::any_of(NumberedInstructions, [](const CodeGenInstruction *Inst) {
return !Inst->HasComplexDeprecationPredicate &&
!Inst->DeprecatedReason.empty();
});
if (HasDeprecationFeatures) {
OS << "extern const uint8_t " << TargetName
<< "InstrDeprecationFeatures[] = {";
Num = 0;
for (const CodeGenInstruction *Inst : NumberedInstructions) {
if (Num % 8 == 0)
OS << "\n ";
if (!Inst->HasComplexDeprecationPredicate &&
!Inst->DeprecatedReason.empty())
OS << Target.getInstNamespace() << "::" << Inst->DeprecatedReason
<< ", ";
else
OS << "uint8_t(-1), ";
++Num;
}
OS << "\n};\n\n";
}
bool HasComplexDeprecationInfos =
llvm::any_of(NumberedInstructions, [](const CodeGenInstruction *Inst) {
return Inst->HasComplexDeprecationPredicate;
});
if (HasComplexDeprecationInfos) {
OS << "extern const MCInstrInfo::ComplexDeprecationPredicate " << TargetName
<< "InstrComplexDeprecationInfos[] = {";
Num = 0;
for (const CodeGenInstruction *Inst : NumberedInstructions) {
if (Num % 8 == 0)
OS << "\n ";
if (Inst->HasComplexDeprecationPredicate)
// Emit a function pointer to the complex predicate method.
OS << "&get" << Inst->DeprecatedReason << "DeprecationInfo, ";
else
OS << "nullptr, ";
++Num;
}
OS << "\n};\n\n";
}
// MCInstrInfo initialization routine.
Records.startTimer("Emit initialization routine");
OS << "static inline void Init" << TargetName
<< "MCInstrInfo(MCInstrInfo *II) {\n";
OS << " II->InitMCInstrInfo(" << TargetName << "Insts, " << TargetName
<< "InstrNameIndices, " << TargetName << "InstrNameData, ";
if (HasDeprecationFeatures)
OS << TargetName << "InstrDeprecationFeatures, ";
else
OS << "nullptr, ";
if (HasComplexDeprecationInfos)
OS << TargetName << "InstrComplexDeprecationInfos, ";
else
OS << "nullptr, ";
OS << NumberedInstructions.size() << ");\n}\n\n";
OS << "} // end namespace llvm\n";
OS << "#endif // GET_INSTRINFO_MC_DESC\n\n";
// Create a TargetInstrInfo subclass to hide the MC layer initialization.
OS << "#ifdef GET_INSTRINFO_HEADER\n";
OS << "#undef GET_INSTRINFO_HEADER\n";
std::string ClassName = TargetName + "GenInstrInfo";
OS << "namespace llvm {\n";
OS << "struct " << ClassName << " : public TargetInstrInfo {\n"
<< " explicit " << ClassName
XRay: Add entry and exit sleds Summary: In this patch we implement the following parts of XRay: - Supporting a function attribute named 'function-instrument' which currently only supports 'xray-always'. We should be able to use this attribute for other instrumentation approaches. - Supporting a function attribute named 'xray-instruction-threshold' used to determine whether a function is instrumented with a minimum number of instructions (IR instruction counts). - X86-specific nop sleds as described in the white paper. - A machine function pass that adds the different instrumentation marker instructions at a very late stage. - A way of identifying which return opcode is considered "normal" for each architecture. There are some caveats here: 1) We don't handle PATCHABLE_RET in platforms other than x86_64 yet -- this means if IR used PATCHABLE_RET directly instead of a normal ret, instruction lowering for that platform might do the wrong thing. We think this should be handled at instruction selection time to by default be unpacked for platforms where XRay is not availble yet. 2) The generated section for X86 is different from what is described from the white paper for the sole reason that LLVM allows us to do this neatly. We're taking the opportunity to deviate from the white paper from this perspective to allow us to get richer information from the runtime library. Reviewers: sanjoy, eugenis, kcc, pcc, echristo, rnk Subscribers: niravd, majnemer, atrick, rnk, emaste, bmakam, mcrosier, mehdi_amini, llvm-commits Differential Revision: http://reviews.llvm.org/D19904 llvm-svn: 275367
2016-07-14 06:06:33 +02:00
<< "(int CFSetupOpcode = -1, int CFDestroyOpcode = -1, int CatchRetOpcode = -1, int ReturnOpcode = -1);\n"
<< " ~" << ClassName << "() override = default;\n";
OS << "\n};\n} // end namespace llvm\n";
OS << "#endif // GET_INSTRINFO_HEADER\n\n";
OS << "#ifdef GET_INSTRINFO_HELPER_DECLS\n";
OS << "#undef GET_INSTRINFO_HELPER_DECLS\n\n";
[tblgen][PredicateExpander] Add the ability to describe more complex constraints on instruction operands. Before this patch, class PredicateExpander only knew how to expand simple predicates that performed checks on instruction operands. In particular, the new scheduling predicate syntax was not rich enough to express checks like this one: Foo(MI->getOperand(0).getImm()) == ExpectedVal; Here, the immediate operand value at index zero is passed in input to function Foo, and ExpectedVal is compared against the value returned by function Foo. While this predicate pattern doesn't show up in any X86 model, it shows up in other upstream targets. So, being able to support those predicates is fundamental if we want to be able to modernize all the scheduling models upstream. With this patch, we allow users to specify if a register/immediate operand value needs to be passed in input to a function as part of the predicate check. Now, register/immediate operand checks all derive from base class CheckOperandBase. This patch also changes where TIIPredicate definitions are expanded by the instructon info emitter. Before, definitions were expanded in class XXXGenInstrInfo (where XXX is a target name). With the introduction of this new syntax, we may want to have TIIPredicates expanded directly in XXXInstrInfo. That is because functions used by the new operand predicates may only exist in the derived class (i.e. XXXInstrInfo). This patch is a non functional change for the existing scheduling models. In future, we will be able to use this richer syntax to better describe complex scheduling predicates, and expose them to llvm-mca. Differential Revision: https://reviews.llvm.org/D53880 llvm-svn: 345714
2018-10-31 13:28:05 +01:00
emitTIIHelperMethods(OS, TargetName, /* ExpandDefintion = */false);
OS << "\n";
OS << "#endif // GET_INSTRINFO_HELPER_DECLS\n\n";
[tblgen][PredicateExpander] Add the ability to describe more complex constraints on instruction operands. Before this patch, class PredicateExpander only knew how to expand simple predicates that performed checks on instruction operands. In particular, the new scheduling predicate syntax was not rich enough to express checks like this one: Foo(MI->getOperand(0).getImm()) == ExpectedVal; Here, the immediate operand value at index zero is passed in input to function Foo, and ExpectedVal is compared against the value returned by function Foo. While this predicate pattern doesn't show up in any X86 model, it shows up in other upstream targets. So, being able to support those predicates is fundamental if we want to be able to modernize all the scheduling models upstream. With this patch, we allow users to specify if a register/immediate operand value needs to be passed in input to a function as part of the predicate check. Now, register/immediate operand checks all derive from base class CheckOperandBase. This patch also changes where TIIPredicate definitions are expanded by the instructon info emitter. Before, definitions were expanded in class XXXGenInstrInfo (where XXX is a target name). With the introduction of this new syntax, we may want to have TIIPredicates expanded directly in XXXInstrInfo. That is because functions used by the new operand predicates may only exist in the derived class (i.e. XXXInstrInfo). This patch is a non functional change for the existing scheduling models. In future, we will be able to use this richer syntax to better describe complex scheduling predicates, and expose them to llvm-mca. Differential Revision: https://reviews.llvm.org/D53880 llvm-svn: 345714
2018-10-31 13:28:05 +01:00
OS << "#ifdef GET_INSTRINFO_HELPERS\n";
OS << "#undef GET_INSTRINFO_HELPERS\n\n";
[tblgen][PredicateExpander] Add the ability to describe more complex constraints on instruction operands. Before this patch, class PredicateExpander only knew how to expand simple predicates that performed checks on instruction operands. In particular, the new scheduling predicate syntax was not rich enough to express checks like this one: Foo(MI->getOperand(0).getImm()) == ExpectedVal; Here, the immediate operand value at index zero is passed in input to function Foo, and ExpectedVal is compared against the value returned by function Foo. While this predicate pattern doesn't show up in any X86 model, it shows up in other upstream targets. So, being able to support those predicates is fundamental if we want to be able to modernize all the scheduling models upstream. With this patch, we allow users to specify if a register/immediate operand value needs to be passed in input to a function as part of the predicate check. Now, register/immediate operand checks all derive from base class CheckOperandBase. This patch also changes where TIIPredicate definitions are expanded by the instructon info emitter. Before, definitions were expanded in class XXXGenInstrInfo (where XXX is a target name). With the introduction of this new syntax, we may want to have TIIPredicates expanded directly in XXXInstrInfo. That is because functions used by the new operand predicates may only exist in the derived class (i.e. XXXInstrInfo). This patch is a non functional change for the existing scheduling models. In future, we will be able to use this richer syntax to better describe complex scheduling predicates, and expose them to llvm-mca. Differential Revision: https://reviews.llvm.org/D53880 llvm-svn: 345714
2018-10-31 13:28:05 +01:00
emitTIIHelperMethods(OS, TargetName, /* ExpandDefintion = */true);
OS << "#endif // GET_INSTRINFO_HELPERS\n\n";
[tblgen][PredicateExpander] Add the ability to describe more complex constraints on instruction operands. Before this patch, class PredicateExpander only knew how to expand simple predicates that performed checks on instruction operands. In particular, the new scheduling predicate syntax was not rich enough to express checks like this one: Foo(MI->getOperand(0).getImm()) == ExpectedVal; Here, the immediate operand value at index zero is passed in input to function Foo, and ExpectedVal is compared against the value returned by function Foo. While this predicate pattern doesn't show up in any X86 model, it shows up in other upstream targets. So, being able to support those predicates is fundamental if we want to be able to modernize all the scheduling models upstream. With this patch, we allow users to specify if a register/immediate operand value needs to be passed in input to a function as part of the predicate check. Now, register/immediate operand checks all derive from base class CheckOperandBase. This patch also changes where TIIPredicate definitions are expanded by the instructon info emitter. Before, definitions were expanded in class XXXGenInstrInfo (where XXX is a target name). With the introduction of this new syntax, we may want to have TIIPredicates expanded directly in XXXInstrInfo. That is because functions used by the new operand predicates may only exist in the derived class (i.e. XXXInstrInfo). This patch is a non functional change for the existing scheduling models. In future, we will be able to use this richer syntax to better describe complex scheduling predicates, and expose them to llvm-mca. Differential Revision: https://reviews.llvm.org/D53880 llvm-svn: 345714
2018-10-31 13:28:05 +01:00
OS << "#ifdef GET_INSTRINFO_CTOR_DTOR\n";
OS << "#undef GET_INSTRINFO_CTOR_DTOR\n";
OS << "namespace llvm {\n";
OS << "extern const MCInstrDesc " << TargetName << "Insts[];\n";
OS << "extern const unsigned " << TargetName << "InstrNameIndices[];\n";
OS << "extern const char " << TargetName << "InstrNameData[];\n";
if (HasDeprecationFeatures)
OS << "extern const uint8_t " << TargetName
<< "InstrDeprecationFeatures[];\n";
if (HasComplexDeprecationInfos)
OS << "extern const MCInstrInfo::ComplexDeprecationPredicate " << TargetName
<< "InstrComplexDeprecationInfos[];\n";
OS << ClassName << "::" << ClassName
<< "(int CFSetupOpcode, int CFDestroyOpcode, int CatchRetOpcode, int "
"ReturnOpcode)\n"
<< " : TargetInstrInfo(CFSetupOpcode, CFDestroyOpcode, CatchRetOpcode, "
"ReturnOpcode) {\n"
<< " InitMCInstrInfo(" << TargetName << "Insts, " << TargetName
<< "InstrNameIndices, " << TargetName << "InstrNameData, ";
if (HasDeprecationFeatures)
OS << TargetName << "InstrDeprecationFeatures, ";
else
OS << "nullptr, ";
if (HasComplexDeprecationInfos)
OS << TargetName << "InstrComplexDeprecationInfos, ";
else
OS << "nullptr, ";
OS << NumberedInstructions.size() << ");\n}\n";
OS << "} // end namespace llvm\n";
OS << "#endif // GET_INSTRINFO_CTOR_DTOR\n\n";
Records.startTimer("Emit operand name mappings");
emitOperandNameMappings(OS, Target, NumberedInstructions);
Records.startTimer("Emit operand type mappings");
emitOperandTypeMappings(OS, Target, NumberedInstructions);
[M68k][TableGen](1/8) TableGen related changes - Add a new TableGen backend: CodeBeads - Add support to generate logical operand information For the first item, it is currently a workaround of M68k's (complex) instruction encoding. A typical architecture, especially CISC one like X86, normally uses `MCInstrDesc::TSFlags` to carry instruction encoding info. However, at the early days of M68k backend development, we found it difficult to fit every possible encoding into the 64-bit `MCInstrDesc::TSFlags`. Therefore CodeBeads was invented to provide an alternative, arbitrary length container for instruciton encoding info. However, in the long term we incline not to use a new TG backend for less common pattern like what we encountered in M68k. A bug has been created to host to discussion on migrating from CodeBeads to more concise solution: https://bugs.llvm.org/show_bug.cgi?id=48792 The second item was also served for similar purpose. It created utility functions that tell you the index of a `MachineOperand` in a `MachineInst` given a logical operand index. In normal cases a logical operand is the same as `MachineOperand`, but for operands using complex addressing mode a logical operand might be consisting of multiple `MachineOperand`. The TableGen-ed `getLogicalOperandIdx`, for instance, can give you the mapping between these two concepts. Nevertheless, we hope to remove this feature in the future if possible. Since it's not really useful for the targets supported by LLVM now either. Authors: myhsu, m4yers, glaubitz Differential Revision: https://reviews.llvm.org/D88385
2021-03-08 01:30:54 +01:00
Records.startTimer("Emit logical operand size mappings");
emitLogicalOperandSizeMappings(OS, TargetName, NumberedInstructions);
Records.startTimer("Emit logical operand type mappings");
emitLogicalOperandTypeMappings(OS, TargetName, NumberedInstructions);
Records.startTimer("Emit helper methods");
emitMCIIHelperMethods(OS, TargetName);
}
void InstrInfoEmitter::emitRecord(const CodeGenInstruction &Inst, unsigned Num,
Record *InstrInfo,
std::map<std::vector<Record*>, unsigned> &EmittedLists,
const OperandInfoMapTy &OpInfo,
raw_ostream &OS) {
int MinOperands = 0;
if (!Inst.Operands.empty())
// Each logical operand can be multiple MI operands.
MinOperands = Inst.Operands.back().MIOperandNo +
Inst.Operands.back().MINumOperands;
OS << " { ";
OS << Num << ",\t" << MinOperands << ",\t"
<< Inst.Operands.NumDefs << ",\t"
<< Inst.TheDef->getValueAsInt("Size") << ",\t"
<< SchedModels.getSchedClassIdx(Inst) << ",\t0";
[MachineOperand][Target] MachineOperand::isRenamable semantics changes Summary: Add a target option AllowRegisterRenaming that is used to opt in to post-register-allocation renaming of registers. This is set to 0 by default, which causes the hasExtraSrcRegAllocReq/hasExtraDstRegAllocReq fields of all opcodes to be set to 1, causing MachineOperand::isRenamable to always return false. Set the AllowRegisterRenaming flag to 1 for all in-tree targets that have lit tests that were effected by enabling COPY forwarding in MachineCopyPropagation (AArch64, AMDGPU, ARM, Hexagon, Mips, PowerPC, RISCV, Sparc, SystemZ and X86). Add some more comments describing the semantics of the MachineOperand::isRenamable function and how it is set and maintained. Change isRenamable to check the operand's opcode hasExtraSrcRegAllocReq/hasExtraDstRegAllocReq bit directly instead of relying on it being consistently reflected in the IsRenamable bit setting. Clear the IsRenamable bit when changing an operand's register value. Remove target code that was clearing the IsRenamable bit when changing registers/opcodes now that this is done conservatively by default. Change setting of hasExtraSrcRegAllocReq in AMDGPU target to be done in one place covering all opcodes that have constant pipe read limit restrictions. Reviewers: qcolombet, MatzeB Subscribers: aemerson, arsenm, jyknight, mcrosier, sdardis, nhaehnle, javed.absar, tpr, arichardson, kristof.beyls, kbarton, fedor.sergeev, asb, rbar, johnrusso, simoncook, jordy.potman.lists, apazos, sabuasal, niosHD, escha, nemanjai, llvm-commits Differential Revision: https://reviews.llvm.org/D43042 llvm-svn: 325931
2018-02-23 19:25:08 +01:00
CodeGenTarget &Target = CDP.getTargetInfo();
// Emit all of the target independent flags...
if (Inst.isPreISelOpcode) OS << "|(1ULL<<MCID::PreISelOpcode)";
if (Inst.isPseudo) OS << "|(1ULL<<MCID::Pseudo)";
if (Inst.isReturn) OS << "|(1ULL<<MCID::Return)";
if (Inst.isEHScopeReturn) OS << "|(1ULL<<MCID::EHScopeReturn)";
if (Inst.isBranch) OS << "|(1ULL<<MCID::Branch)";
if (Inst.isIndirectBranch) OS << "|(1ULL<<MCID::IndirectBranch)";
if (Inst.isCompare) OS << "|(1ULL<<MCID::Compare)";
if (Inst.isMoveImm) OS << "|(1ULL<<MCID::MoveImm)";
if (Inst.isMoveReg) OS << "|(1ULL<<MCID::MoveReg)";
if (Inst.isBitcast) OS << "|(1ULL<<MCID::Bitcast)";
if (Inst.isAdd) OS << "|(1ULL<<MCID::Add)";
if (Inst.isTrap) OS << "|(1ULL<<MCID::Trap)";
if (Inst.isSelect) OS << "|(1ULL<<MCID::Select)";
if (Inst.isBarrier) OS << "|(1ULL<<MCID::Barrier)";
if (Inst.hasDelaySlot) OS << "|(1ULL<<MCID::DelaySlot)";
if (Inst.isCall) OS << "|(1ULL<<MCID::Call)";
if (Inst.canFoldAsLoad) OS << "|(1ULL<<MCID::FoldableAsLoad)";
if (Inst.mayLoad) OS << "|(1ULL<<MCID::MayLoad)";
if (Inst.mayStore) OS << "|(1ULL<<MCID::MayStore)";
Allow target to handle STRICT floating-point nodes The ISD::STRICT_ nodes used to implement the constrained floating-point intrinsics are currently never passed to the target back-end, which makes it impossible to handle them correctly (e.g. mark instructions are depending on a floating-point status and control register, or mark instructions as possibly trapping). This patch allows the target to use setOperationAction to switch the action on ISD::STRICT_ nodes to Legal. If this is done, the SelectionDAG common code will stop converting the STRICT nodes to regular floating-point nodes, but instead pass the STRICT nodes to the target using normal SelectionDAG matching rules. To avoid having the back-end duplicate all the floating-point instruction patterns to handle both strict and non-strict variants, we make the MI codegen explicitly aware of the floating-point exceptions by introducing two new concepts: - A new MCID flag "mayRaiseFPException" that the target should set on any instruction that possibly can raise FP exception according to the architecture definition. - A new MI flag FPExcept that CodeGen/SelectionDAG will set on any MI instruction resulting from expansion of any constrained FP intrinsic. Any MI instruction that is *both* marked as mayRaiseFPException *and* FPExcept then needs to be considered as raising exceptions by MI-level codegen (e.g. scheduling). Setting those two new flags is straightforward. The mayRaiseFPException flag is simply set via TableGen by marking all relevant instruction patterns in the .td files. The FPExcept flag is set in SDNodeFlags when creating the STRICT_ nodes in the SelectionDAG, and gets inherited in the MachineSDNode nodes created from it during instruction selection. The flag is then transfered to an MIFlag when creating the MI from the MachineSDNode. This is handled just like fast-math flags like no-nans are handled today. This patch includes both common code changes required to implement the new features, and the SystemZ implementation. Reviewed By: andrew.w.kaylor Differential Revision: https://reviews.llvm.org/D55506 llvm-svn: 362663
2019-06-06 00:33:10 +02:00
if (Inst.mayRaiseFPException) OS << "|(1ULL<<MCID::MayRaiseFPException)";
if (Inst.isPredicable) OS << "|(1ULL<<MCID::Predicable)";
if (Inst.isConvertibleToThreeAddress) OS << "|(1ULL<<MCID::ConvertibleTo3Addr)";
if (Inst.isCommutable) OS << "|(1ULL<<MCID::Commutable)";
if (Inst.isTerminator) OS << "|(1ULL<<MCID::Terminator)";
if (Inst.isReMaterializable) OS << "|(1ULL<<MCID::Rematerializable)";
if (Inst.isNotDuplicable) OS << "|(1ULL<<MCID::NotDuplicable)";
if (Inst.Operands.hasOptionalDef) OS << "|(1ULL<<MCID::HasOptionalDef)";
if (Inst.usesCustomInserter) OS << "|(1ULL<<MCID::UsesCustomInserter)";
if (Inst.hasPostISelHook) OS << "|(1ULL<<MCID::HasPostISelHook)";
if (Inst.Operands.isVariadic)OS << "|(1ULL<<MCID::Variadic)";
if (Inst.hasSideEffects) OS << "|(1ULL<<MCID::UnmodeledSideEffects)";
if (Inst.isAsCheapAsAMove) OS << "|(1ULL<<MCID::CheapAsAMove)";
[MachineOperand][Target] MachineOperand::isRenamable semantics changes Summary: Add a target option AllowRegisterRenaming that is used to opt in to post-register-allocation renaming of registers. This is set to 0 by default, which causes the hasExtraSrcRegAllocReq/hasExtraDstRegAllocReq fields of all opcodes to be set to 1, causing MachineOperand::isRenamable to always return false. Set the AllowRegisterRenaming flag to 1 for all in-tree targets that have lit tests that were effected by enabling COPY forwarding in MachineCopyPropagation (AArch64, AMDGPU, ARM, Hexagon, Mips, PowerPC, RISCV, Sparc, SystemZ and X86). Add some more comments describing the semantics of the MachineOperand::isRenamable function and how it is set and maintained. Change isRenamable to check the operand's opcode hasExtraSrcRegAllocReq/hasExtraDstRegAllocReq bit directly instead of relying on it being consistently reflected in the IsRenamable bit setting. Clear the IsRenamable bit when changing an operand's register value. Remove target code that was clearing the IsRenamable bit when changing registers/opcodes now that this is done conservatively by default. Change setting of hasExtraSrcRegAllocReq in AMDGPU target to be done in one place covering all opcodes that have constant pipe read limit restrictions. Reviewers: qcolombet, MatzeB Subscribers: aemerson, arsenm, jyknight, mcrosier, sdardis, nhaehnle, javed.absar, tpr, arichardson, kristof.beyls, kbarton, fedor.sergeev, asb, rbar, johnrusso, simoncook, jordy.potman.lists, apazos, sabuasal, niosHD, escha, nemanjai, llvm-commits Differential Revision: https://reviews.llvm.org/D43042 llvm-svn: 325931
2018-02-23 19:25:08 +01:00
if (!Target.getAllowRegisterRenaming() || Inst.hasExtraSrcRegAllocReq)
OS << "|(1ULL<<MCID::ExtraSrcRegAllocReq)";
if (!Target.getAllowRegisterRenaming() || Inst.hasExtraDefRegAllocReq)
OS << "|(1ULL<<MCID::ExtraDefRegAllocReq)";
if (Inst.isRegSequence) OS << "|(1ULL<<MCID::RegSequence)";
if (Inst.isExtractSubreg) OS << "|(1ULL<<MCID::ExtractSubreg)";
if (Inst.isInsertSubreg) OS << "|(1ULL<<MCID::InsertSubreg)";
if (Inst.isConvergent) OS << "|(1ULL<<MCID::Convergent)";
if (Inst.variadicOpsAreDefs) OS << "|(1ULL<<MCID::VariadicOpsAreDefs)";
if (Inst.isAuthenticated) OS << "|(1ULL<<MCID::Authenticated)";
// Emit all of the target-specific flags...
BitsInit *TSF = Inst.TheDef->getValueAsBitsInit("TSFlags");
if (!TSF)
PrintFatalError(Inst.TheDef->getLoc(), "no TSFlags?");
uint64_t Value = 0;
for (unsigned i = 0, e = TSF->getNumBits(); i != e; ++i) {
if (const auto *Bit = dyn_cast<BitInit>(TSF->getBit(i)))
Value |= uint64_t(Bit->getValue()) << i;
else
PrintFatalError(Inst.TheDef->getLoc(),
"Invalid TSFlags bit in " + Inst.TheDef->getName());
}
OS << ", 0x";
OS.write_hex(Value);
2010-06-09 18:16:48 +02:00
OS << "ULL, ";
// Emit the implicit uses and defs lists...
std::vector<Record*> UseList = Inst.TheDef->getValueAsListOfDefs("Uses");
if (UseList.empty())
OS << "nullptr, ";
else
OS << "ImplicitList" << EmittedLists[UseList] << ", ";
std::vector<Record*> DefList = Inst.TheDef->getValueAsListOfDefs("Defs");
if (DefList.empty())
OS << "nullptr, ";
else
OS << "ImplicitList" << EmittedLists[DefList] << ", ";
// Emit the operand info.
std::vector<std::string> OperandInfo = GetOperandInfo(Inst);
if (OperandInfo.empty())
OS << "nullptr";
else
OS << "OperandInfo" << OpInfo.find(OperandInfo)->second;
OS << " }, // Inst #" << Num << " = " << Inst.TheDef->getName() << "\n";
}
// emitEnums - Print out enum values for all of the instructions.
void InstrInfoEmitter::emitEnums(raw_ostream &OS) {
OS << "#ifdef GET_INSTRINFO_ENUM\n";
OS << "#undef GET_INSTRINFO_ENUM\n";
OS << "namespace llvm {\n\n";
const CodeGenTarget &Target = CDP.getTargetInfo();
// We must emit the PHI opcode first...
StringRef Namespace = Target.getInstNamespace();
2012-04-11 23:02:30 +02:00
if (Namespace.empty())
PrintFatalError("No instructions defined!");
OS << "namespace " << Namespace << " {\n";
OS << " enum {\n";
unsigned Num = 0;
for (const CodeGenInstruction *Inst : Target.getInstructionsByEnumValue())
OS << " " << Inst->TheDef->getName() << "\t= " << Num++ << ",\n";
OS << " INSTRUCTION_LIST_END = " << Num << "\n";
OS << " };\n\n";
OS << "} // end namespace " << Namespace << "\n";
OS << "} // end namespace llvm\n";
OS << "#endif // GET_INSTRINFO_ENUM\n\n";
OS << "#ifdef GET_INSTRINFO_SCHED_ENUM\n";
OS << "#undef GET_INSTRINFO_SCHED_ENUM\n";
OS << "namespace llvm {\n\n";
OS << "namespace " << Namespace << " {\n";
OS << "namespace Sched {\n";
OS << " enum {\n";
Num = 0;
for (const auto &Class : SchedModels.explicit_classes())
OS << " " << Class.Name << "\t= " << Num++ << ",\n";
OS << " SCHED_LIST_END = " << Num << "\n";
OS << " };\n";
OS << "} // end namespace Sched\n";
OS << "} // end namespace " << Namespace << "\n";
OS << "} // end namespace llvm\n";
OS << "#endif // GET_INSTRINFO_SCHED_ENUM\n\n";
}
namespace llvm {
void EmitInstrInfo(RecordKeeper &RK, raw_ostream &OS) {
RK.startTimer("Analyze DAG patterns");
InstrInfoEmitter(RK).run(OS);
RK.startTimer("Emit map table");
EmitMapTable(RK, OS);
}
} // end namespace llvm