llvm-project/mlir/lib/TableGen/Operator.cpp

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

578 lines
20 KiB
C++
Raw Normal View History

//===- Operator.cpp - Operator class --------------------------------------===//
//
// 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
//
//===----------------------------------------------------------------------===//
//
// Operator wrapper to simplify using TableGen Record defining a MLIR Op.
//
//===----------------------------------------------------------------------===//
#include "mlir/TableGen/Operator.h"
#include "mlir/TableGen/OpTrait.h"
#include "mlir/TableGen/Predicate.h"
#include "mlir/TableGen/Type.h"
#include "llvm/ADT/EquivalenceClasses.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/Sequence.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/StringExtras.h"
#include "llvm/ADT/TypeSwitch.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/FormatVariadic.h"
#include "llvm/TableGen/Error.h"
#include "llvm/TableGen/Record.h"
#define DEBUG_TYPE "mlir-tblgen-operator"
using namespace mlir;
using namespace mlir::tblgen;
using llvm::DagInit;
using llvm::DefInit;
using llvm::Record;
Operator::Operator(const llvm::Record &def)
: dialect(def.getValueAsDef("opDialect")), def(def) {
// The first `_` in the op's TableGen def name is treated as separating the
// dialect prefix and the op class name. The dialect prefix will be ignored if
// not empty. Otherwise, if def name starts with a `_`, the `_` is considered
// as part of the class name.
StringRef prefix;
std::tie(prefix, cppClassName) = def.getName().split('_');
if (prefix.empty()) {
// Class name with a leading underscore and without dialect prefix
cppClassName = def.getName();
} else if (cppClassName.empty()) {
// Class name without dialect prefix
cppClassName = prefix;
}
populateOpStructure();
}
std::string Operator::getOperationName() const {
auto prefix = dialect.getName();
auto opName = def.getValueAsString("opName");
if (prefix.empty())
return std::string(opName);
return std::string(llvm::formatv("{0}.{1}", prefix, opName));
}
std::string Operator::getAdaptorName() const {
return std::string(llvm::formatv("{0}Adaptor", getCppClassName()));
}
StringRef Operator::getDialectName() const { return dialect.getName(); }
StringRef Operator::getCppClassName() const { return cppClassName; }
std::string Operator::getQualCppClassName() const {
auto prefix = dialect.getCppNamespace();
if (prefix.empty())
return std::string(cppClassName);
return std::string(llvm::formatv("{0}::{1}", prefix, cppClassName));
}
int Operator::getNumResults() const {
DagInit *results = def.getValueAsDag("results");
return results->getNumArgs();
}
StringRef Operator::getExtraClassDeclaration() const {
constexpr auto attr = "extraClassDeclaration";
if (def.isValueUnset(attr))
return {};
return def.getValueAsString(attr);
}
const llvm::Record &Operator::getDef() const { return def; }
bool Operator::skipDefaultBuilders() const {
return def.getValueAsBit("skipDefaultBuilders");
}
auto Operator::result_begin() -> value_iterator { return results.begin(); }
auto Operator::result_end() -> value_iterator { return results.end(); }
auto Operator::getResults() -> value_range {
return {result_begin(), result_end()};
}
TypeConstraint Operator::getResultTypeConstraint(int index) const {
DagInit *results = def.getValueAsDag("results");
return TypeConstraint(cast<DefInit>(results->getArg(index)));
}
StringRef Operator::getResultName(int index) const {
DagInit *results = def.getValueAsDag("results");
return results->getArgNameStr(index);
}
auto Operator::getResultDecorators(int index) const -> var_decorator_range {
Record *result =
cast<DefInit>(def.getValueAsDag("results")->getArg(index))->getDef();
if (!result->isSubClassOf("OpVariable"))
return var_decorator_range(nullptr, nullptr);
return *result->getValueAsListInit("decorators");
}
unsigned Operator::getNumVariableLengthResults() const {
return llvm::count_if(results, [](const NamedTypeConstraint &c) {
return c.constraint.isVariableLength();
});
}
unsigned Operator::getNumVariableLengthOperands() const {
return llvm::count_if(operands, [](const NamedTypeConstraint &c) {
return c.constraint.isVariableLength();
});
}
bool Operator::hasSingleVariadicArg() const {
return getNumArgs() == 1 && getArg(0).is<NamedTypeConstraint *>() &&
getOperand(0).isVariadic();
}
Operator::arg_iterator Operator::arg_begin() const { return arguments.begin(); }
Operator::arg_iterator Operator::arg_end() const { return arguments.end(); }
Operator::arg_range Operator::getArgs() const {
return {arg_begin(), arg_end()};
}
StringRef Operator::getArgName(int index) const {
DagInit *argumentValues = def.getValueAsDag("arguments");
[mlir] Add basic support for attributes in ODS-generated Python bindings In ODS, attributes of an operation can be provided as a part of the "arguments" field, together with operands. Such attributes are accepted by the op builder and have accessors generated. Implement similar functionality for ODS-generated op-specific Python bindings: the `__init__` method now accepts arguments together with operands, in the same order as in the ODS `arguments` field; the instance properties are introduced to OpView classes to access the attributes. This initial implementation accepts and returns instances of the corresponding attribute class, and not the underlying values since the mapping scheme of the value types between C++, C and Python is not yet clear. Default-valued attributes are not supported as that would require Python to be able to parse C++ literals. Since attributes in ODS are tightely related to the actual C++ type system, provide a separate Tablegen file with the mapping between ODS storage type for attributes (typically, the underlying C++ attribute class), and the corresponding class name. So far, this might look unnecessary since all names match exactly, but this is not necessarily the cases for non-standard, out-of-tree attributes, which may also be placed in non-default namespaces or Python modules. This also allows out-of-tree users to generate Python bindings without having to modify the bindings generator itself. Storage type was preferred over the Tablegen "def" of the attribute class because ODS essentially encodes attribute _constraints_ rather than classes, e.g. there may be many Tablegen "def"s in the ODS that correspond to the same attribute type with additional constraints The presence of the explicit mapping requires the change in the .td file structure: instead of just calling the bindings generator directly on the main ODS file of the dialect, it becomes necessary to create a new file that includes the main ODS file of the dialect and provides the mapping for attribute types. Arguably, this approach offers better separability of the Python bindings in the build system as the main dialect no longer needs to know that it is being processed by the bindings generator. Reviewed By: stellaraccident Differential Revision: https://reviews.llvm.org/D91542
2020-11-16 23:17:03 +08:00
return argumentValues->getArgNameStr(index);
}
auto Operator::getArgDecorators(int index) const -> var_decorator_range {
Record *arg =
cast<DefInit>(def.getValueAsDag("arguments")->getArg(index))->getDef();
if (!arg->isSubClassOf("OpVariable"))
return var_decorator_range(nullptr, nullptr);
return *arg->getValueAsListInit("decorators");
}
const OpTrait *Operator::getTrait(StringRef trait) const {
for (const auto &t : traits) {
if (const auto *opTrait = dyn_cast<NativeOpTrait>(&t)) {
if (opTrait->getTrait() == trait)
return opTrait;
} else if (const auto *opTrait = dyn_cast<InternalOpTrait>(&t)) {
if (opTrait->getTrait() == trait)
return opTrait;
} else if (const auto *opTrait = dyn_cast<InterfaceOpTrait>(&t)) {
if (opTrait->getTrait() == trait)
return opTrait;
}
}
return nullptr;
}
auto Operator::region_begin() const -> const_region_iterator {
return regions.begin();
}
auto Operator::region_end() const -> const_region_iterator {
return regions.end();
}
auto Operator::getRegions() const
-> llvm::iterator_range<const_region_iterator> {
return {region_begin(), region_end()};
}
unsigned Operator::getNumRegions() const { return regions.size(); }
const NamedRegion &Operator::getRegion(unsigned index) const {
return regions[index];
}
unsigned Operator::getNumVariadicRegions() const {
return llvm::count_if(regions,
[](const NamedRegion &c) { return c.isVariadic(); });
}
auto Operator::successor_begin() const -> const_successor_iterator {
return successors.begin();
}
auto Operator::successor_end() const -> const_successor_iterator {
return successors.end();
}
auto Operator::getSuccessors() const
-> llvm::iterator_range<const_successor_iterator> {
return {successor_begin(), successor_end()};
}
unsigned Operator::getNumSuccessors() const { return successors.size(); }
const NamedSuccessor &Operator::getSuccessor(unsigned index) const {
return successors[index];
}
unsigned Operator::getNumVariadicSuccessors() const {
return llvm::count_if(successors,
[](const NamedSuccessor &c) { return c.isVariadic(); });
}
auto Operator::trait_begin() const -> const_trait_iterator {
return traits.begin();
}
auto Operator::trait_end() const -> const_trait_iterator {
return traits.end();
}
auto Operator::getTraits() const -> llvm::iterator_range<const_trait_iterator> {
return {trait_begin(), trait_end()};
}
auto Operator::attribute_begin() const -> attribute_iterator {
return attributes.begin();
}
auto Operator::attribute_end() const -> attribute_iterator {
return attributes.end();
}
auto Operator::getAttributes() const
-> llvm::iterator_range<attribute_iterator> {
return {attribute_begin(), attribute_end()};
}
auto Operator::operand_begin() -> value_iterator { return operands.begin(); }
auto Operator::operand_end() -> value_iterator { return operands.end(); }
auto Operator::getOperands() -> value_range {
return {operand_begin(), operand_end()};
}
auto Operator::getArg(int index) const -> Argument { return arguments[index]; }
// Mapping from result index to combined argument and result index. Arguments
// are indexed to match getArg index, while the result indexes are mapped to
// avoid overlap.
static int resultIndex(int i) { return -1 - i; }
bool Operator::isVariadic() const {
return any_of(llvm::concat<const NamedTypeConstraint>(operands, results),
[](const NamedTypeConstraint &op) { return op.isVariadic(); });
}
void Operator::populateTypeInferenceInfo(
const llvm::StringMap<int> &argumentsAndResultsIndex) {
// If the type inference op interface is not registered, then do not attempt
// to determine if the result types an be inferred.
auto &recordKeeper = def.getRecords();
auto *inferTrait = recordKeeper.getDef(inferTypeOpInterface);
allResultsHaveKnownTypes = false;
if (!inferTrait)
return;
// If there are no results, the skip this else the build method generated
// overlaps with another autogenerated builder.
if (getNumResults() == 0)
return;
// Skip for ops with variadic operands/results.
// TODO: This can be relaxed.
if (isVariadic())
return;
// Skip cases currently being custom generated.
// TODO: Remove special cases.
if (getTrait("::mlir::OpTrait::SameOperandsAndResultType"))
return;
// We create equivalence classes of argument/result types where arguments
// and results are mapped into the same index space and indices corresponding
// to the same type are in the same equivalence class.
llvm::EquivalenceClasses<int> ecs;
resultTypeMapping.resize(getNumResults());
// Captures the argument whose type matches a given result type. Preference
// towards capturing operands first before attributes.
auto captureMapping = [&](int i) {
bool found = false;
ecs.insert(resultIndex(i));
auto mi = ecs.findLeader(resultIndex(i));
for (auto me = ecs.member_end(); mi != me; ++mi) {
if (*mi < 0) {
auto tc = getResultTypeConstraint(i);
if (tc.getBuilderCall().hasValue()) {
resultTypeMapping[i].emplace_back(tc);
found = true;
}
continue;
}
if (getArg(*mi).is<NamedAttribute *>()) {
// TODO: Handle attributes.
continue;
} else {
resultTypeMapping[i].emplace_back(*mi);
found = true;
}
}
return found;
};
for (const OpTrait &trait : traits) {
const llvm::Record &def = trait.getDef();
// If the infer type op interface was manually added, then treat it as
// intention that the op needs special handling.
// TODO: Reconsider whether to always generate, this is more conservative
// and keeps existing behavior so starting that way for now.
if (def.isSubClassOf(
llvm::formatv("{0}::Trait", inferTypeOpInterface).str()))
return;
if (const auto *opTrait = dyn_cast<InterfaceOpTrait>(&trait))
if (&opTrait->getDef() == inferTrait)
return;
if (!def.isSubClassOf("AllTypesMatch"))
continue;
auto values = def.getValueAsListOfStrings("values");
auto root = argumentsAndResultsIndex.lookup(values.front());
for (StringRef str : values)
ecs.unionSets(argumentsAndResultsIndex.lookup(str), root);
}
// Verifies that all output types have a corresponding known input type
// and chooses matching operand or attribute (in that order) that
// matches it.
allResultsHaveKnownTypes =
all_of(llvm::seq<int>(0, getNumResults()), captureMapping);
// If the types could be computed, then add type inference trait.
if (allResultsHaveKnownTypes)
traits.push_back(OpTrait::create(inferTrait->getDefInit()));
}
void Operator::populateOpStructure() {
auto &recordKeeper = def.getRecords();
auto *typeConstraintClass = recordKeeper.getClass("TypeConstraint");
auto *attrClass = recordKeeper.getClass("Attr");
auto *derivedAttrClass = recordKeeper.getClass("DerivedAttr");
auto *opVarClass = recordKeeper.getClass("OpVariable");
numNativeAttributes = 0;
DagInit *argumentValues = def.getValueAsDag("arguments");
unsigned numArgs = argumentValues->getNumArgs();
// Mapping from name of to argument or result index. Arguments are indexed
// to match getArg index, while the results are negatively indexed.
llvm::StringMap<int> argumentsAndResultsIndex;
// Handle operands and native attributes.
for (unsigned i = 0; i != numArgs; ++i) {
auto *arg = argumentValues->getArg(i);
auto givenName = argumentValues->getArgNameStr(i);
auto *argDefInit = dyn_cast<DefInit>(arg);
if (!argDefInit)
PrintFatalError(def.getLoc(),
Twine("undefined type for argument #") + Twine(i));
Record *argDef = argDefInit->getDef();
if (argDef->isSubClassOf(opVarClass))
argDef = argDef->getValueAsDef("constraint");
if (argDef->isSubClassOf(typeConstraintClass)) {
operands.push_back(
NamedTypeConstraint{givenName, TypeConstraint(argDef)});
} else if (argDef->isSubClassOf(attrClass)) {
if (givenName.empty())
PrintFatalError(argDef->getLoc(), "attributes must be named");
if (argDef->isSubClassOf(derivedAttrClass))
PrintFatalError(argDef->getLoc(),
"derived attributes not allowed in argument list");
attributes.push_back({givenName, Attribute(argDef)});
++numNativeAttributes;
} else {
PrintFatalError(def.getLoc(), "unexpected def type; only defs deriving "
"from TypeConstraint or Attr are allowed");
}
if (!givenName.empty())
argumentsAndResultsIndex[givenName] = i;
}
// Handle derived attributes.
for (const auto &val : def.getValues()) {
if (auto *record = dyn_cast<llvm::RecordRecTy>(val.getType())) {
if (!record->isSubClassOf(attrClass))
continue;
if (!record->isSubClassOf(derivedAttrClass))
PrintFatalError(def.getLoc(),
"unexpected Attr where only DerivedAttr is allowed");
if (record->getClasses().size() != 1) {
PrintFatalError(
def.getLoc(),
"unsupported attribute modelling, only single class expected");
}
attributes.push_back(
{cast<llvm::StringInit>(val.getNameInit())->getValue(),
Attribute(cast<DefInit>(val.getValue()))});
}
}
// Populate `arguments`. This must happen after we've finalized `operands` and
// `attributes` because we will put their elements' pointers in `arguments`.
// SmallVector may perform re-allocation under the hood when adding new
// elements.
int operandIndex = 0, attrIndex = 0;
for (unsigned i = 0; i != numArgs; ++i) {
Record *argDef = dyn_cast<DefInit>(argumentValues->getArg(i))->getDef();
if (argDef->isSubClassOf(opVarClass))
argDef = argDef->getValueAsDef("constraint");
if (argDef->isSubClassOf(typeConstraintClass)) {
attrOrOperandMapping.push_back(
{OperandOrAttribute::Kind::Operand, operandIndex});
arguments.emplace_back(&operands[operandIndex++]);
} else {
assert(argDef->isSubClassOf(attrClass));
attrOrOperandMapping.push_back(
{OperandOrAttribute::Kind::Attribute, attrIndex});
arguments.emplace_back(&attributes[attrIndex++]);
}
}
auto *resultsDag = def.getValueAsDag("results");
auto *outsOp = dyn_cast<DefInit>(resultsDag->getOperator());
if (!outsOp || outsOp->getDef()->getName() != "outs") {
PrintFatalError(def.getLoc(), "'results' must have 'outs' directive");
}
// Handle results.
for (unsigned i = 0, e = resultsDag->getNumArgs(); i < e; ++i) {
auto name = resultsDag->getArgNameStr(i);
auto *resultInit = dyn_cast<DefInit>(resultsDag->getArg(i));
if (!resultInit) {
PrintFatalError(def.getLoc(),
Twine("undefined type for result #") + Twine(i));
}
auto *resultDef = resultInit->getDef();
if (resultDef->isSubClassOf(opVarClass))
resultDef = resultDef->getValueAsDef("constraint");
results.push_back({name, TypeConstraint(resultDef)});
if (!name.empty())
argumentsAndResultsIndex[name] = resultIndex(i);
}
// Handle successors
auto *successorsDag = def.getValueAsDag("successors");
auto *successorsOp = dyn_cast<DefInit>(successorsDag->getOperator());
if (!successorsOp || successorsOp->getDef()->getName() != "successor") {
PrintFatalError(def.getLoc(),
"'successors' must have 'successor' directive");
}
for (unsigned i = 0, e = successorsDag->getNumArgs(); i < e; ++i) {
auto name = successorsDag->getArgNameStr(i);
auto *successorInit = dyn_cast<DefInit>(successorsDag->getArg(i));
if (!successorInit) {
PrintFatalError(def.getLoc(),
Twine("undefined kind for successor #") + Twine(i));
}
Successor successor(successorInit->getDef());
// Only support variadic successors if it is the last one for now.
if (i != e - 1 && successor.isVariadic())
PrintFatalError(def.getLoc(), "only the last successor can be variadic");
successors.push_back({name, successor});
}
// Create list of traits, skipping over duplicates: appending to lists in
// tablegen is easy, making them unique less so, so dedupe here.
if (auto *traitList = def.getValueAsListInit("traits")) {
// This is uniquing based on pointers of the trait.
SmallPtrSet<const llvm::Init *, 32> traitSet;
traits.reserve(traitSet.size());
for (auto *traitInit : *traitList) {
// Keep traits in the same order while skipping over duplicates.
if (traitSet.insert(traitInit).second)
traits.push_back(OpTrait::create(traitInit));
}
}
populateTypeInferenceInfo(argumentsAndResultsIndex);
// Handle regions
auto *regionsDag = def.getValueAsDag("regions");
auto *regionsOp = dyn_cast<DefInit>(regionsDag->getOperator());
if (!regionsOp || regionsOp->getDef()->getName() != "region") {
PrintFatalError(def.getLoc(), "'regions' must have 'region' directive");
}
for (unsigned i = 0, e = regionsDag->getNumArgs(); i < e; ++i) {
auto name = regionsDag->getArgNameStr(i);
auto *regionInit = dyn_cast<DefInit>(regionsDag->getArg(i));
if (!regionInit) {
PrintFatalError(def.getLoc(),
Twine("undefined kind for region #") + Twine(i));
}
Region region(regionInit->getDef());
if (region.isVariadic()) {
// Only support variadic regions if it is the last one for now.
if (i != e - 1)
PrintFatalError(def.getLoc(), "only the last region can be variadic");
if (name.empty())
PrintFatalError(def.getLoc(), "variadic regions must be named");
}
regions.push_back({name, region});
}
LLVM_DEBUG(print(llvm::dbgs()));
}
auto Operator::getSameTypeAsResult(int index) const -> ArrayRef<ArgOrType> {
assert(allResultTypesKnown());
return resultTypeMapping[index];
}
ArrayRef<llvm::SMLoc> Operator::getLoc() const { return def.getLoc(); }
bool Operator::hasDescription() const {
return def.getValue("description") != nullptr;
}
StringRef Operator::getDescription() const {
return def.getValueAsString("description");
}
bool Operator::hasSummary() const { return def.getValue("summary") != nullptr; }
StringRef Operator::getSummary() const {
return def.getValueAsString("summary");
}
bool Operator::hasAssemblyFormat() const {
auto *valueInit = def.getValueInit("assemblyFormat");
return isa<llvm::CodeInit, llvm::StringInit>(valueInit);
}
StringRef Operator::getAssemblyFormat() const {
return TypeSwitch<llvm::Init *, StringRef>(def.getValueInit("assemblyFormat"))
.Case<llvm::StringInit, llvm::CodeInit>(
[&](auto *init) { return init->getValue(); });
}
void Operator::print(llvm::raw_ostream &os) const {
os << "op '" << getOperationName() << "'\n";
for (Argument arg : arguments) {
if (auto *attr = arg.dyn_cast<NamedAttribute *>())
os << "[attribute] " << attr->name << '\n';
else
os << "[operand] " << arg.get<NamedTypeConstraint *>()->name << '\n';
}
}
auto Operator::VariableDecoratorIterator::unwrap(llvm::Init *init)
-> VariableDecorator {
return VariableDecorator(cast<llvm::DefInit>(init)->getDef());
}
auto Operator::getArgToOperandOrAttribute(int index) const
-> OperandOrAttribute {
return attrOrOperandMapping[index];
}