Index: include/llvm/CodeGen/GlobalISel/LegalizerInfo.h =================================================================== --- include/llvm/CodeGen/GlobalISel/LegalizerInfo.h +++ include/llvm/CodeGen/GlobalISel/LegalizerInfo.h @@ -21,6 +21,7 @@ #include "llvm/ADT/STLExtras.h" #include "llvm/ADT/SmallVector.h" #include "llvm/CodeGen/TargetOpcodes.h" +#include "llvm/Support/raw_ostream.h" #include "llvm/Support/LowLevelTypeImpl.h" #include #include @@ -79,6 +80,10 @@ /// Sentinel value for when no action was found in the specified table. NotFound, + + /// Fall back onto the old rules. + /// TODO: Remove this once we've migrated + UseLegacyRules, }; /// Legalization is decided based on an instruction's opcode, which type slot @@ -104,6 +109,271 @@ LegalityQuery(unsigned Opcode, ArrayRef Types) : Opcode(Opcode), Types(Types) {} + + raw_ostream &print(raw_ostream &OS) const; +}; + +using LegalityPredicate = std::function; +using LegalizeMutation = + std::function(const LegalityQuery &)>; + +namespace LegalityPredicates { +LegalityPredicate all(LegalityPredicate P0, LegalityPredicate P1); +LegalityPredicate typeInSet(unsigned TypeIdx, + std::initializer_list TypesInit); +LegalityPredicate +typePairInSet(unsigned TypeIdx0, unsigned TypeIdx1, + std::initializer_list> TypesInit); +LegalityPredicate isScalar(unsigned TypeIdx); +LegalityPredicate narrowerThan(unsigned TypeIdx, unsigned Size); +LegalityPredicate widerThan(unsigned TypeIdx, unsigned Size); +LegalityPredicate sizeNotPow2(unsigned TypeIdx); +LegalityPredicate numElementsNotPow2(unsigned TypeIdx); +} // end namespace LegalityPredicates + +namespace LegalizeMutations { +LegalizeMutation pick(unsigned TypeIdx, LLT Ty); +LegalizeMutation widenScalarToPow2(unsigned TypeIdx, unsigned Min = 0); +LegalizeMutation moreElementsToPow2(unsigned TypeIdx, unsigned Min = 0); +} // end namespace LegalizeMutations + +class LegalizeRule { + LegalityPredicate Predicate; + LegalizeAction Action; + LegalizeMutation Mutation; + +public: + LegalizeRule(LegalityPredicate Predicate, LegalizeAction Action, + LegalizeMutation Mutation = nullptr) + : Predicate(Predicate), Action(Action), Mutation(Mutation) {} + + bool match(const LegalityQuery &Query) const { + return Predicate(Query); + } + + LegalizeAction getAction() const { return Action; } + std::pair determineMutation(const LegalityQuery &Query) const { + if (Mutation) + return Mutation(Query); + return std::make_pair(0, LLT{}); + } +}; + +using LegalizerActionTuple = std::tuple; + +class LegalizeRuleSet { + SmallVector Rules; + + void add(const LegalizeRule &Rule) { + Rules.push_back(Rule); + } + + static bool always(const LegalityQuery &) { return true; } + + inline LegalizeRuleSet &actionIf(LegalizeAction Action, + LegalityPredicate Predicate) { + add({Predicate, Action}); + return *this; + } + inline LegalizeRuleSet &actionIf(LegalizeAction Action, + LegalityPredicate Predicate, + LegalizeMutation Mutation) { + add({Predicate, Action, Mutation}); + return *this; + } + inline LegalizeRuleSet &actionFor(LegalizeAction Action, + std::initializer_list Types) { + return actionIf(Action, LegalityPredicates::typeInSet(0, Types)); + } + inline LegalizeRuleSet & + actionFor(LegalizeAction Action, + std::initializer_list> Types) { + return actionIf(Action, LegalityPredicates::typePairInSet(0, 1, Types)); + } + inline LegalizeRuleSet & + actionForCartesianProd(LegalizeAction Action, + std::initializer_list Types) { + return actionIf(Action, LegalityPredicates::all( + LegalityPredicates::typeInSet(0, Types), + LegalityPredicates::typeInSet(1, Types))); + } + inline LegalizeRuleSet & + actionForCartesianProd(LegalizeAction Action, + std::initializer_list Types0, + std::initializer_list Types1) { + return actionIf(Action, LegalityPredicates::all( + LegalityPredicates::typeInSet(0, Types0), + LegalityPredicates::typeInSet(1, Types1))); + } + +public: + // Primitive rule definition functions + LegalizeRuleSet &legalIf(LegalityPredicate Predicate) { + return actionIf(LegalizeAction::Legal, Predicate); + } + LegalizeRuleSet &legalFor(std::initializer_list Types) { + return actionFor(LegalizeAction::Legal, Types); + } + LegalizeRuleSet &legalFor(std::initializer_list> Types) { + return actionFor(LegalizeAction::Legal, Types); + } + LegalizeRuleSet &legalForCartesianProd(std::initializer_list Types) { + return actionForCartesianProd(LegalizeAction::Legal, Types); + } + LegalizeRuleSet &legalForCartesianProd(std::initializer_list Types0, + std::initializer_list Types1) { + return actionForCartesianProd(LegalizeAction::Legal, Types0, Types1); + } + + LegalizeRuleSet &libcallIf(LegalityPredicate Predicate) { + return actionIf(LegalizeAction::Libcall, Predicate); + } + LegalizeRuleSet &libcallFor(std::initializer_list Types) { + return actionFor(LegalizeAction::Libcall, Types); + } + LegalizeRuleSet & + libcallFor(std::initializer_list> Types) { + return actionFor(LegalizeAction::Libcall, Types); + } + LegalizeRuleSet &libcallForCartesianProd(std::initializer_list Types) { + return actionForCartesianProd(LegalizeAction::Libcall, Types); + } + LegalizeRuleSet &libcallForCartesianProd(std::initializer_list Types0, + std::initializer_list Types1) { + return actionForCartesianProd(LegalizeAction::Libcall, Types0, Types1); + } + +#if 0 + LegalizeRuleSet &lowerFor(std::initializer_list Types) { + return lowerIf(LegalityPredicates::typeInSet(0, Types)); + } + + LegalizeRuleSet &lowerFor(std::initializer_list> Types) { + return lowerIf(LegalityPredicates::typePairInSet(0, 1, Types)); + } +#endif + + LegalizeRuleSet &widenScalarIf(LegalityPredicate Predicate, + LegalizeMutation Mutation) { + return actionIf(LegalizeAction::WidenScalar, Predicate, Mutation); + } + LegalizeRuleSet &narrowScalarIf(LegalityPredicate Predicate, + LegalizeMutation Mutation) { + return actionIf(LegalizeAction::NarrowScalar, Predicate, Mutation); + } + + LegalizeRuleSet &moreElementsIf(LegalityPredicate Predicate, + LegalizeMutation Mutation) { + return actionIf(LegalizeAction::MoreElements, Predicate, Mutation); + } + LegalizeRuleSet &fewerElementsIf(LegalityPredicate Predicate, + LegalizeMutation Mutation) { + return actionIf(LegalizeAction::FewerElements, Predicate, Mutation); + } + + LegalizeRuleSet &unsupported() { + return actionIf(LegalizeAction::Unsupported, always); + } + LegalizeRuleSet &unsupportedIf(LegalityPredicate Predicate) { + return actionIf(LegalizeAction::Unsupported, Predicate); + } + + LegalizeRuleSet &customIf(LegalityPredicate Predicate) { + return actionIf(LegalizeAction::Custom, Predicate); + } + LegalizeRuleSet &customForCartesianProd(std::initializer_list Types) { + return actionForCartesianProd(LegalizeAction::Custom, Types); + } + LegalizeRuleSet &customForCartesianProd(std::initializer_list Types0, + std::initializer_list Types1) { + return actionForCartesianProd(LegalizeAction::Custom, Types0, Types1); + } + + // Convenience rule definition functions + LegalizeRuleSet &widenScalarToPow2(unsigned TypeIdx, unsigned MinSize = 0) { + return widenScalarIf( + LegalityPredicates::sizeNotPow2(TypeIdx), + LegalizeMutations::widenScalarToPow2(TypeIdx, MinSize)); + } + + LegalizeRuleSet &narrowScalar(unsigned TypeIdx, LegalizeMutation Mutation) { + return narrowScalarIf(LegalityPredicates::isScalar(TypeIdx), Mutation); + } + + LegalizeRuleSet &minScalar(unsigned TypeIdx, const LLT &Ty) { + return widenScalarIf( + LegalityPredicates::narrowerThan(TypeIdx, Ty.getSizeInBits()), + LegalizeMutations::pick(TypeIdx, Ty)); + } + + LegalizeRuleSet &maxScalar(unsigned TypeIdx, const LLT &Ty) { + return narrowScalarIf( + LegalityPredicates::widerThan(TypeIdx, Ty.getSizeInBits()), + LegalizeMutations::pick(TypeIdx, Ty)); + } + + LegalizeRuleSet &maxScalarIf(LegalityPredicate Predicate, unsigned TypeIdx, const LLT &Ty) { + return narrowScalarIf( + [=](const LegalityQuery &Query) { + return LegalityPredicates::widerThan(TypeIdx, Ty.getSizeInBits()) && + Predicate(Query); + }, + LegalizeMutations::pick(TypeIdx, Ty)); + } + + LegalizeRuleSet &clampScalar(unsigned TypeIdx, const LLT &MinTy, const LLT &MaxTy) { + assert(MinTy.isScalar() && MaxTy.isScalar() && "Expected scalar types"); + + return minScalar(TypeIdx, MinTy) + .maxScalar(TypeIdx, MaxTy); + } + + LegalizeRuleSet &moreElementsToPow2(unsigned TypeIdx) { + return moreElementsIf(LegalityPredicates::numElementsNotPow2(TypeIdx), + LegalizeMutations::moreElementsToPow2(TypeIdx)); + } + + LegalizeRuleSet &minNumElements(unsigned TypeIdx, const LLT &EltTy, unsigned MinElements) { + return moreElementsIf( + [=](const LegalityQuery &Query) { + LLT VecTy = Query.Types[TypeIdx]; + return VecTy.getElementType() == EltTy && + VecTy.getNumElements() < MinElements; + }, + [=](const LegalityQuery &Query) { + LLT VecTy = Query.Types[TypeIdx]; + return std::make_pair( + TypeIdx, LLT::vector(MinElements, VecTy.getScalarSizeInBits())); + }); + } + LegalizeRuleSet &maxNumElements(unsigned TypeIdx, const LLT &EltTy, unsigned MaxElements) { + return fewerElementsIf( + [=](const LegalityQuery &Query) { + LLT VecTy = Query.Types[TypeIdx]; + return VecTy.getElementType() == EltTy && + VecTy.getNumElements() > MaxElements; + }, + [=](const LegalityQuery &Query) { + LLT VecTy = Query.Types[TypeIdx]; + return std::make_pair( + TypeIdx, LLT::vector(MaxElements, VecTy.getScalarSizeInBits())); + }); + } + LegalizeRuleSet &clampNumElements(unsigned TypeIdx, const LLT &MinTy, const LLT &MaxTy) { + assert(MinTy.getElementType() == MaxTy.getElementType() && + "Expected element types to agree"); + + const LLT &EltTy = MinTy.getElementType(); + return minNumElements(TypeIdx, EltTy, MinTy.getNumElements()) + .maxNumElements(TypeIdx, EltTy, MaxTy.getNumElements()); + } + + LegalizeRuleSet &fallback() { + add({always, LegalizeAction::UseLegacyRules}); + return *this; + } + + LegalizerActionTuple apply(const LegalityQuery &Query) const; }; class LegalizerInfo { @@ -267,13 +537,15 @@ LegalizeAction DecreaseAction, LegalizeAction IncreaseAction); + LegalizeRuleSet &getActionDefinitions(unsigned Opcode); + const LegalizeRuleSet &getActionDefinitions(unsigned Opcode) const; + /// Determine what action should be taken to legalize the described /// instruction. Requires computeTables to have been called. /// /// \returns a pair consisting of the kind of legalization that should be /// performed and the destination type. - std::tuple - getAction(const LegalityQuery &Query) const; + LegalizerActionTuple getAction(const LegalityQuery &Query) const; /// Determine what action should be taken to legalize the given generic /// instruction. @@ -281,8 +553,8 @@ /// \returns a tuple consisting of the LegalizeAction that should be /// performed, the type-index it should be performed on and the destination /// type. - std::tuple - getAction(const MachineInstr &MI, const MachineRegisterInfo &MRI) const; + LegalizerActionTuple getAction(const MachineInstr &MI, + const MachineRegisterInfo &MRI) const; bool isLegal(const MachineInstr &MI, const MachineRegisterInfo &MRI) const; @@ -473,6 +745,8 @@ AddrSpace2PointerActions[LastOp - FirstOp + 1]; std::unordered_map> NumElements2Actions[LastOp - FirstOp + 1]; + + LegalizeRuleSet RulesForOpcode[LastOp - FirstOp + 1]; }; } // end namespace llvm. Index: lib/CodeGen/GlobalISel/CMakeLists.txt =================================================================== --- lib/CodeGen/GlobalISel/CMakeLists.txt +++ lib/CodeGen/GlobalISel/CMakeLists.txt @@ -4,8 +4,10 @@ IRTranslator.cpp InstructionSelect.cpp InstructionSelector.cpp - LegalizerHelper.cpp + LegalityPredicates.cpp + LegalizeMutations.cpp Legalizer.cpp + LegalizerHelper.cpp LegalizerInfo.cpp Localizer.cpp MachineIRBuilder.cpp Index: lib/CodeGen/GlobalISel/LegalityPredicates.cpp =================================================================== --- /dev/null +++ lib/CodeGen/GlobalISel/LegalityPredicates.cpp @@ -0,0 +1,78 @@ +//===- lib/CodeGen/GlobalISel/LegalizerPredicates.cpp - Predicates --------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// A library of predicate factories to use for LegalityPredicate. +// +//===----------------------------------------------------------------------===// + +#include "llvm/CodeGen/GlobalISel/LegalizerInfo.h" + +using namespace llvm; + +LegalityPredicate +LegalityPredicates::all(LegalityPredicate P0, LegalityPredicate P1) { + return [=](const LegalityQuery &Query) { + return P0(Query) && P1(Query); + }; +} + +LegalityPredicate +LegalityPredicates::typeInSet(unsigned TypeIdx, + std::initializer_list TypesInit) { + SmallVector Types = TypesInit; + return [=](const LegalityQuery &Query) { + return std::find(Types.begin(), Types.end(), Query.Types[TypeIdx]) != Types.end(); + }; +} + +LegalityPredicate LegalityPredicates::typePairInSet( + unsigned TypeIdx0, unsigned TypeIdx1, + std::initializer_list> TypesInit) { + SmallVector, 4> Types = TypesInit; + return [=](const LegalityQuery &Query) { + std::pair Match = {Query.Types[TypeIdx0], Query.Types[TypeIdx1]}; + return std::find(Types.begin(), Types.end(), Match) != Types.end(); + }; +} + +LegalityPredicate LegalityPredicates::isScalar(unsigned TypeIdx) { + return [=](const LegalityQuery &Query) { + return Query.Types[TypeIdx].isScalar(); + }; +} + +LegalityPredicate LegalityPredicates::narrowerThan(unsigned TypeIdx, + unsigned Size) { + return [=](const LegalityQuery &Query) { + const LLT &QueryTy = Query.Types[TypeIdx]; + return QueryTy.isScalar() && QueryTy.getSizeInBits() < Size; + }; +} + +LegalityPredicate LegalityPredicates::widerThan(unsigned TypeIdx, + unsigned Size) { + return [=](const LegalityQuery &Query) { + const LLT &QueryTy = Query.Types[TypeIdx]; + return QueryTy.isScalar() && QueryTy.getSizeInBits() > Size; + }; +} + +LegalityPredicate LegalityPredicates::sizeNotPow2(unsigned TypeIdx) { + return [=](const LegalityQuery &Query) { + const LLT &QueryTy = Query.Types[TypeIdx]; + return QueryTy.isScalar() && !isPowerOf2_32(QueryTy.getSizeInBits()); + }; +} + +LegalityPredicate LegalityPredicates::numElementsNotPow2(unsigned TypeIdx) { + return [=](const LegalityQuery &Query) { + const LLT &QueryTy = Query.Types[TypeIdx]; + return QueryTy.isVector() && isPowerOf2_32(QueryTy.getNumElements()); + }; +} Index: lib/CodeGen/GlobalISel/LegalizeMutations.cpp =================================================================== --- /dev/null +++ lib/CodeGen/GlobalISel/LegalizeMutations.cpp @@ -0,0 +1,44 @@ +//===- lib/CodeGen/GlobalISel/LegalizerMutations.cpp - Mutations ----------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// A library of mutation factories to use for LegalityMutation. +// +//===----------------------------------------------------------------------===// + +#include "llvm/CodeGen/GlobalISel/LegalizerInfo.h" + +using namespace llvm; + +LegalizeMutation LegalizeMutations::pick(unsigned TypeIdx, LLT Ty) { + return + [=](const LegalityQuery &Query) { return std::make_pair(TypeIdx, Ty); }; +} + +LegalizeMutation LegalizeMutations::widenScalarToPow2(unsigned TypeIdx, + unsigned Min) { + return [=](const LegalityQuery &Query) { + unsigned NewSizeInBits = + 1 << Log2_32_Ceil(Query.Types[TypeIdx].getSizeInBits()); + if (NewSizeInBits < Min) + NewSizeInBits = Min; + return std::make_pair(TypeIdx, LLT::scalar(NewSizeInBits)); + }; +} + +LegalizeMutation LegalizeMutations::moreElementsToPow2(unsigned TypeIdx, + unsigned Min) { + return [=](const LegalityQuery &Query) { + const LLT &VecTy = Query.Types[TypeIdx]; + unsigned NewNumElements = 1 << Log2_32_Ceil(VecTy.getNumElements()); + if (NewNumElements < Min) + NewNumElements = Min; + return std::make_pair( + TypeIdx, LLT::vector(NewNumElements, VecTy.getScalarSizeInBits())); + }; +} Index: lib/CodeGen/GlobalISel/LegalizerInfo.cpp =================================================================== --- lib/CodeGen/GlobalISel/LegalizerInfo.cpp +++ lib/CodeGen/GlobalISel/LegalizerInfo.cpp @@ -24,6 +24,7 @@ #include "llvm/CodeGen/MachineRegisterInfo.h" #include "llvm/CodeGen/TargetOpcodes.h" #include "llvm/MC/MCInstrDesc.h" +#include "llvm/Support/Debug.h" #include "llvm/Support/ErrorHandling.h" #include "llvm/Support/LowLevelTypeImpl.h" #include "llvm/Support/MathExtras.h" @@ -31,6 +32,35 @@ #include using namespace llvm; +#define DEBUG_TYPE "legalizer-info" + +raw_ostream &LegalityQuery::print(raw_ostream &OS) const { + OS << Opcode << ", {"; + for (const auto &Type : Types) { + OS << Type << ", "; + } + OS << "}"; + return OS; +} + +LegalizerActionTuple LegalizeRuleSet::apply(const LegalityQuery &Query) const { + DEBUG(dbgs() << "Applying legalizer ruleset to: "; Query.print(dbgs()); dbgs() << "\n"); + for (const auto &Rule : Rules) { + if (Rule.match(Query)) { + DEBUG(dbgs() << ".. match\n"); + std::pair Mutation = Rule.determineMutation(Query); + DEBUG(dbgs() << ".. .. " << (unsigned)Rule.getAction() << ", " + << Mutation.first << ", " << Mutation.second << "\n"); + assert(Query.Types[Mutation.first] != Mutation.second && + "Simple loop detected"); + return std::make_tuple(Rule.getAction(), Mutation.first, Mutation.second); + } else + DEBUG(dbgs() << ".. no match\n"); + } + DEBUG(dbgs() << ".. fallback to legacy rules\n"); + return std::make_tuple(LegalizeAction::UseLegacyRules, 0, LLT{}); +} + LegalizerInfo::LegalizerInfo() : TablesInitialized(false) { // Set defaults. // FIXME: these two (G_ANYEXT and G_TRUNC?) can be legalized to the @@ -186,17 +216,40 @@ return MRI.getType(MI.getOperand(OpIdx).getReg()); } -std::tuple +LegalizeRuleSet &LegalizerInfo::getActionDefinitions(unsigned Opcode) { + assert(Opcode >= FirstOp && Opcode <= LastOp && "Unsupported opcode"); + const unsigned OpcodeIdx = Opcode - FirstOp; + return RulesForOpcode[OpcodeIdx]; +} + +const LegalizeRuleSet & +LegalizerInfo::getActionDefinitions(unsigned Opcode) const { + assert(Opcode >= FirstOp && Opcode <= LastOp && "Unsupported opcode"); + const unsigned OpcodeIdx = Opcode - FirstOp; + return RulesForOpcode[OpcodeIdx]; +} + +LegalizerActionTuple LegalizerInfo::getAction(const LegalityQuery &Query) const { + LegalizerActionTuple Action = getActionDefinitions(Query.Opcode).apply(Query); + if (std::get<0>(Action) != LegalizeAction::UseLegacyRules) { + return Action; + } + for (unsigned i = 0; i < Query.Types.size(); ++i) { auto Action = getAspectAction({Query.Opcode, i, Query.Types[i]}); - if (Action.first != LegalizeAction::Legal) + if (Action.first != LegalizeAction::Legal) { + DEBUG(dbgs() << ".. (legacy) Type " << i << " Action=" + << (unsigned)Action.first << ", " << Action.second << "\n"); return std::make_tuple(Action.first, i, Action.second); + } else + DEBUG(dbgs() << ".. (legacy) Type " << i << " Legal\n"); } + DEBUG(dbgs() << ".. (legacy) Legal\n"); return std::make_tuple(LegalizeAction::Legal, 0, LLT{}); } -std::tuple +LegalizerActionTuple LegalizerInfo::getAction(const MachineInstr &MI, const MachineRegisterInfo &MRI) const { SmallVector Types; @@ -321,6 +374,7 @@ case LegalizeAction::Unsupported: return {Size, LegalizeAction::Unsupported}; case LegalizeAction::NotFound: + case LegalizeAction::UseLegacyRules: llvm_unreachable("NotFound"); } llvm_unreachable("Action has an unknown enum value"); Index: lib/Target/AArch64/AArch64LegalizerInfo.cpp =================================================================== --- lib/Target/AArch64/AArch64LegalizerInfo.cpp +++ lib/Target/AArch64/AArch64LegalizerInfo.cpp @@ -23,110 +23,7 @@ #include "llvm/IR/Type.h" using namespace llvm; - -/// FIXME: The following static functions are SizeChangeStrategy functions -/// that are meant to temporarily mimic the behaviour of the old legalization -/// based on doubling/halving non-legal types as closely as possible. This is -/// not entirly possible as only legalizing the types that are exactly a power -/// of 2 times the size of the legal types would require specifying all those -/// sizes explicitly. -/// In practice, not specifying those isn't a problem, and the below functions -/// should disappear quickly as we add support for legalizing non-power-of-2 -/// sized types further. -static void -addAndInterleaveWithUnsupported(LegalizerInfo::SizeAndActionsVec &result, - const LegalizerInfo::SizeAndActionsVec &v) { - for (unsigned i = 0; i < v.size(); ++i) { - result.push_back(v[i]); - if (i + 1 < v[i].first && i + 1 < v.size() && - v[i + 1].first != v[i].first + 1) - result.push_back({v[i].first + 1, LegalizeAction::Unsupported}); - } -} - -static LegalizerInfo::SizeAndActionsVec -widen_1_narrow_128_ToLargest(const LegalizerInfo::SizeAndActionsVec &v) { - assert(v.size() >= 1); - assert(v[0].first > 2); - LegalizerInfo::SizeAndActionsVec result = {{1, LegalizeAction::WidenScalar}, - {2, LegalizeAction::Unsupported}}; - addAndInterleaveWithUnsupported(result, v); - auto Largest = result.back().first; - assert(Largest + 1 < 128); - result.push_back({Largest + 1, LegalizeAction::Unsupported}); - result.push_back({128, LegalizeAction::NarrowScalar}); - result.push_back({129, LegalizeAction::Unsupported}); - return result; -} - -static LegalizerInfo::SizeAndActionsVec -widen_16(const LegalizerInfo::SizeAndActionsVec &v) { - assert(v.size() >= 1); - assert(v[0].first > 17); - LegalizerInfo::SizeAndActionsVec result = {{1, LegalizeAction::Unsupported}, - {16, LegalizeAction::WidenScalar}, - {17, LegalizeAction::Unsupported}}; - addAndInterleaveWithUnsupported(result, v); - auto Largest = result.back().first; - result.push_back({Largest + 1, LegalizeAction::Unsupported}); - return result; -} - -static LegalizerInfo::SizeAndActionsVec -widen_1_8(const LegalizerInfo::SizeAndActionsVec &v) { - assert(v.size() >= 1); - assert(v[0].first > 9); - LegalizerInfo::SizeAndActionsVec result = { - {1, LegalizeAction::WidenScalar}, {2, LegalizeAction::Unsupported}, - {8, LegalizeAction::WidenScalar}, {9, LegalizeAction::Unsupported}}; - addAndInterleaveWithUnsupported(result, v); - auto Largest = result.back().first; - result.push_back({Largest + 1, LegalizeAction::Unsupported}); - return result; -} - -static LegalizerInfo::SizeAndActionsVec -widen_1_8_16(const LegalizerInfo::SizeAndActionsVec &v) { - assert(v.size() >= 1); - assert(v[0].first > 17); - LegalizerInfo::SizeAndActionsVec result = { - {1, LegalizeAction::WidenScalar}, {2, LegalizeAction::Unsupported}, - {8, LegalizeAction::WidenScalar}, {9, LegalizeAction::Unsupported}, - {16, LegalizeAction::WidenScalar}, {17, LegalizeAction::Unsupported}}; - addAndInterleaveWithUnsupported(result, v); - auto Largest = result.back().first; - result.push_back({Largest + 1, LegalizeAction::Unsupported}); - return result; -} - -static LegalizerInfo::SizeAndActionsVec -widen_1_8_16_narrowToLargest(const LegalizerInfo::SizeAndActionsVec &v) { - assert(v.size() >= 1); - assert(v[0].first > 17); - LegalizerInfo::SizeAndActionsVec result = { - {1, LegalizeAction::WidenScalar}, {2, LegalizeAction::Unsupported}, - {8, LegalizeAction::WidenScalar}, {9, LegalizeAction::Unsupported}, - {16, LegalizeAction::WidenScalar}, {17, LegalizeAction::Unsupported}}; - addAndInterleaveWithUnsupported(result, v); - auto Largest = result.back().first; - result.push_back({Largest + 1, LegalizeAction::NarrowScalar}); - return result; -} - -static LegalizerInfo::SizeAndActionsVec -widen_1_8_16_32(const LegalizerInfo::SizeAndActionsVec &v) { - assert(v.size() >= 1); - assert(v[0].first > 33); - LegalizerInfo::SizeAndActionsVec result = { - {1, LegalizeAction::WidenScalar}, {2, LegalizeAction::Unsupported}, - {8, LegalizeAction::WidenScalar}, {9, LegalizeAction::Unsupported}, - {16, LegalizeAction::WidenScalar}, {17, LegalizeAction::Unsupported}, - {32, LegalizeAction::WidenScalar}, {33, LegalizeAction::Unsupported}}; - addAndInterleaveWithUnsupported(result, v); - auto Largest = result.back().first; - result.push_back({Largest + 1, LegalizeAction::Unsupported}); - return result; -} +using namespace LegalityPredicates; AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST) { using namespace TargetOpcode; @@ -137,44 +34,60 @@ const LLT s32 = LLT::scalar(32); const LLT s64 = LLT::scalar(64); const LLT s128 = LLT::scalar(128); + const LLT s256 = LLT::scalar(256); + const LLT s512 = LLT::scalar(512); + const LLT v16s8 = LLT::vector(16, 8); + const LLT v8s8 = LLT::vector(8, 8); + const LLT v4s8 = LLT::vector(4, 8); + const LLT v8s16 = LLT::vector(8, 16); + const LLT v4s16 = LLT::vector(4, 16); + const LLT v2s16 = LLT::vector(2, 16); const LLT v2s32 = LLT::vector(2, 32); const LLT v4s32 = LLT::vector(4, 32); const LLT v2s64 = LLT::vector(2, 64); - for (auto Ty : {p0, s1, s8, s16, s32, s64}) - setAction({G_IMPLICIT_DEF, Ty}, LegalizeAction::Legal); + getActionDefinitions(G_IMPLICIT_DEF) + .legalFor({p0, s1, s8, s16, s32, s64}) + .clampScalar(0, s1, s64) + .widenScalarToPow2(0, 8) + .unsupported(); - for (auto Ty : {s16, s32, s64, p0}) - setAction({G_PHI, Ty}, LegalizeAction::Legal); + getActionDefinitions(G_PHI) + .legalFor({p0, s16, s32, s64}) + .clampScalar(0, s16, s64) + .widenScalarToPow2(0) + .unsupported(); - setLegalizeScalarToDifferentSizeStrategy(G_PHI, 0, widen_1_8); - - for (auto Ty : { s32, s64 }) - setAction({G_BSWAP, Ty}, LegalizeAction::Legal); + getActionDefinitions(G_BSWAP) + .legalFor({s32, s64}) + .unsupported(); for (unsigned BinOp : {G_ADD, G_SUB, G_MUL, G_AND, G_OR, G_XOR, G_SHL}) { - // These operations naturally get the right answer when used on - // GPR32, even if the actual type is narrower. - for (auto Ty : {s32, s64, v2s32, v4s32, v2s64}) - setAction({BinOp, Ty}, LegalizeAction::Legal); - - if (BinOp != G_ADD) - setLegalizeScalarToDifferentSizeStrategy(BinOp, 0, - widen_1_8_16_narrowToLargest); + getActionDefinitions(BinOp) + .legalFor({s32, s64, v2s32, v4s32, v2s64}) + .clampScalar(0, s32, s64) + .widenScalarToPow2(0) + .clampNumElements(0, v2s32, v4s32) + .clampNumElements(0, v2s64, v2s64) + .moreElementsToPow2(0) + .unsupported(); } - setAction({G_GEP, p0}, LegalizeAction::Legal); - setAction({G_GEP, 1, s64}, LegalizeAction::Legal); - - setLegalizeScalarToDifferentSizeStrategy(G_GEP, 1, widen_1_8_16_32); + getActionDefinitions(G_GEP) + .legalFor({{p0, s64}}) + .clampScalar(1, s64, s64) + .unsupported(); - setAction({G_PTR_MASK, p0}, LegalizeAction::Legal); + getActionDefinitions(G_PTR_MASK) + .legalFor({p0}) + .unsupported(); for (unsigned BinOp : {G_LSHR, G_ASHR, G_SDIV, G_UDIV}) { - for (auto Ty : {s32, s64}) - setAction({BinOp, Ty}, LegalizeAction::Legal); - - setLegalizeScalarToDifferentSizeStrategy(BinOp, 0, widen_1_8_16); + getActionDefinitions(BinOp) + .legalFor({s32, s64}) + .clampScalar(0, s32, s64) + .widenScalarToPow2(0) + .unsupported(); } for (unsigned BinOp : {G_SREM, G_UREM}) @@ -186,204 +99,298 @@ setAction({Op, 1, s1}, LegalizeAction::Legal); } - for (unsigned Op : {G_UADDE, G_USUBE, G_SADDO, G_SSUBO, G_SMULH, G_UMULH}) { - for (auto Ty : { s32, s64 }) - setAction({Op, Ty}, LegalizeAction::Legal); - - setAction({Op, 1, s1}, LegalizeAction::Legal); + for (unsigned Op : {G_SMULH, G_UMULH}) { + getActionDefinitions(Op) + .legalFor({s32, s64}) + .unsupported(); + } + for (unsigned Op : {G_UADDE, G_USUBE, G_SADDO, G_SSUBO}) { + getActionDefinitions(Op) + .legalFor({{s32, s1}, {s64, s1}}) + .unsupported(); } for (unsigned BinOp : {G_FADD, G_FSUB, G_FMA, G_FMUL, G_FDIV}) - for (auto Ty : {s32, s64}) - setAction({BinOp, Ty}, LegalizeAction::Legal); + getActionDefinitions(BinOp).legalFor({s32, s64}).unsupported(); for (unsigned BinOp : {G_FREM, G_FPOW}) { - setAction({BinOp, s32}, LegalizeAction::Libcall); - setAction({BinOp, s64}, LegalizeAction::Libcall); - } - - for (auto Ty : {s32, s64, p0}) { - setAction({G_INSERT, Ty}, LegalizeAction::Legal); - setAction({G_INSERT, 1, Ty}, LegalizeAction::Legal); + getActionDefinitions(BinOp).libcallFor({s32, s64}).unsupported(); } - setLegalizeScalarToDifferentSizeStrategy(G_INSERT, 0, - widen_1_8_16_narrowToLargest); - for (auto Ty : {s1, s8, s16}) { - setAction({G_INSERT, 1, Ty}, LegalizeAction::Legal); - // FIXME: Can't widen the sources because that violates the constraints on - // G_INSERT (It seems entirely reasonable that inputs shouldn't overlap). - } - - for (auto Ty : {s1, s8, s16, s32, s64, p0}) - setAction({G_EXTRACT, Ty}, LegalizeAction::Legal); - for (auto Ty : {s32, s64}) - setAction({G_EXTRACT, 1, Ty}, LegalizeAction::Legal); + getActionDefinitions(G_INSERT) + .unsupportedIf([=](const LegalityQuery &Query) { + return Query.Types[0].getSizeInBits() <= Query.Types[1].getSizeInBits(); + }) + .legalIf([=](const LegalityQuery &Query) { + const LLT &Ty0 = Query.Types[0]; + const LLT &Ty1 = Query.Types[1]; + if (Ty0 != s32 && Ty0 != s64 && Ty0 != p0) + return false; + return isPowerOf2_32(Ty1.getSizeInBits()) && + (Ty1.getSizeInBits() == 1 || Ty1.getSizeInBits() >= 8); + }) + .clampScalar(0, s32, s64) + .widenScalarToPow2(0) + .maxScalarIf(typeInSet(0, {s32}), 1, s16) + .maxScalarIf(typeInSet(0, {s64}), 1, s32) + .widenScalarToPow2(1) + .unsupported(); + + getActionDefinitions(G_EXTRACT) + .unsupportedIf([=](const LegalityQuery &Query) { + return Query.Types[0].getSizeInBits() >= Query.Types[1].getSizeInBits(); + }) + .legalIf([=](const LegalityQuery &Query) { + const LLT &Ty0 = Query.Types[0]; + const LLT &Ty1 = Query.Types[1]; + if (Ty1 != s32 && Ty1 != s64) + return false; + if (Ty1 == p0) + return true; + return isPowerOf2_32(Ty0.getSizeInBits()) && + (Ty0.getSizeInBits() == 1 || Ty0.getSizeInBits() >= 8); + }) + .clampScalar(1, s32, s64) + .widenScalarToPow2(1) + .maxScalarIf(typeInSet(1, {s32}), 0, s16) + .maxScalarIf(typeInSet(1, {s64}), 0, s32) + .widenScalarToPow2(0) + .unsupported(); for (unsigned MemOp : {G_LOAD, G_STORE}) { - for (auto Ty : {s8, s16, s32, s64, p0, v2s32}) - setAction({MemOp, Ty}, LegalizeAction::Legal); - - setLegalizeScalarToDifferentSizeStrategy(MemOp, 0, - widen_1_narrow_128_ToLargest); - - // And everything's fine in addrspace 0. - setAction({MemOp, 1, p0}, LegalizeAction::Legal); + getActionDefinitions(MemOp) + .legalFor({{s8, p0}, {s16, p0}, {s32, p0}, {s64, p0}, {p0, p0}, {v2s32, p0}}) + .clampScalar(0, s8, s64) + .widenScalarToPow2(0) + .clampNumElements(0, v2s32, v2s32) + .unsupported(); } // Constants - for (auto Ty : {s32, s64}) { - setAction({TargetOpcode::G_CONSTANT, Ty}, LegalizeAction::Legal); - setAction({TargetOpcode::G_FCONSTANT, Ty}, LegalizeAction::Legal); - } - - setAction({G_CONSTANT, p0}, LegalizeAction::Legal); - - setLegalizeScalarToDifferentSizeStrategy(G_CONSTANT, 0, widen_1_8_16); - setLegalizeScalarToDifferentSizeStrategy(G_FCONSTANT, 0, widen_16); - - setAction({G_ICMP, 1, s32}, LegalizeAction::Legal); - setAction({G_ICMP, 1, s64}, LegalizeAction::Legal); - setAction({G_ICMP, 1, p0}, LegalizeAction::Legal); - - setLegalizeScalarToDifferentSizeStrategy(G_ICMP, 0, widen_1_8_16); - setLegalizeScalarToDifferentSizeStrategy(G_FCMP, 0, widen_1_8_16); - setLegalizeScalarToDifferentSizeStrategy(G_ICMP, 1, widen_1_8_16); - - setAction({G_ICMP, s32}, LegalizeAction::Legal); - setAction({G_FCMP, s32}, LegalizeAction::Legal); - setAction({G_FCMP, 1, s32}, LegalizeAction::Legal); - setAction({G_FCMP, 1, s64}, LegalizeAction::Legal); + getActionDefinitions(G_CONSTANT) + .legalFor({p0, s32, s64}) + .clampScalar(0, s32, s64) + .widenScalarToPow2(0) + .unsupported(); + getActionDefinitions(G_FCONSTANT) + .legalFor({s32, s64}) + .clampScalar(0, s32, s64) + .unsupported(); + + getActionDefinitions(G_ICMP) + .legalFor({{s32, s32}, {s32, s64}, {s32, p0}}) + .clampScalar(0, s32, s32) + .clampScalar(1, s32, s64) + .widenScalarToPow2(1) + .unsupported(); + + getActionDefinitions(G_FCMP) + .legalFor({{s32, s32}, {s32, s64}}) + .clampScalar(0, s32, s32) + .clampScalar(1, s32, s64) + .widenScalarToPow2(1) + .unsupported(); // Extensions - for (auto Ty : { s1, s8, s16, s32, s64 }) { - setAction({G_ZEXT, Ty}, LegalizeAction::Legal); - setAction({G_SEXT, Ty}, LegalizeAction::Legal); - setAction({G_ANYEXT, Ty}, LegalizeAction::Legal); + for (unsigned Op : {G_ZEXT, G_SEXT, G_ANYEXT}) { + getActionDefinitions(Op) + .legalFor({s1, s8, s16, s32, s64}) + .maxScalar(0, s64) + .widenScalarToPow2(0) + .unsupported(); } // FP conversions - for (auto Ty : { s16, s32 }) { - setAction({G_FPTRUNC, Ty}, LegalizeAction::Legal); - setAction({G_FPEXT, 1, Ty}, LegalizeAction::Legal); - } - - for (auto Ty : { s32, s64 }) { - setAction({G_FPTRUNC, 1, Ty}, LegalizeAction::Legal); - setAction({G_FPEXT, Ty}, LegalizeAction::Legal); - } + getActionDefinitions(G_FPTRUNC) + .legalFor({{s16, s32}, {s16, s64}, {s32, s64}}) + .unsupported(); + getActionDefinitions(G_FPEXT) + .legalFor({{s32, s16}, {s64, s16}, {s64, s32}}) + .unsupported(); // Conversions - for (auto Ty : { s32, s64 }) { - setAction({G_FPTOSI, 0, Ty}, LegalizeAction::Legal); - setAction({G_FPTOUI, 0, Ty}, LegalizeAction::Legal); - setAction({G_SITOFP, 1, Ty}, LegalizeAction::Legal); - setAction({G_UITOFP, 1, Ty}, LegalizeAction::Legal); - } - setLegalizeScalarToDifferentSizeStrategy(G_FPTOSI, 0, widen_1_8_16); - setLegalizeScalarToDifferentSizeStrategy(G_FPTOUI, 0, widen_1_8_16); - setLegalizeScalarToDifferentSizeStrategy(G_SITOFP, 1, widen_1_8_16); - setLegalizeScalarToDifferentSizeStrategy(G_UITOFP, 1, widen_1_8_16); - - for (auto Ty : { s32, s64 }) { - setAction({G_FPTOSI, 1, Ty}, LegalizeAction::Legal); - setAction({G_FPTOUI, 1, Ty}, LegalizeAction::Legal); - setAction({G_SITOFP, 0, Ty}, LegalizeAction::Legal); - setAction({G_UITOFP, 0, Ty}, LegalizeAction::Legal); - } + for (unsigned Op : {G_FPTOSI, G_FPTOUI}) + getActionDefinitions(Op) + .legalForCartesianProd({s32, s64}) + .clampScalar(0, s32, s64) + .widenScalarToPow2(0) + .clampScalar(1, s32, s64) + .widenScalarToPow2(1) + .unsupported(); + + for (unsigned Op : {G_SITOFP, G_UITOFP}) + getActionDefinitions(Op) + .legalForCartesianProd({s32, s64}) + .clampScalar(1, s32, s64) + .widenScalarToPow2(1) + .clampScalar(0, s32, s64) + .widenScalarToPow2(0) + .unsupported(); // Control-flow - for (auto Ty : {s1, s8, s16, s32}) - setAction({G_BRCOND, Ty}, LegalizeAction::Legal); - setAction({G_BRINDIRECT, p0}, LegalizeAction::Legal); + getActionDefinitions(G_BRCOND).legalFor({s1, s8, s16, s32}).unsupported(); + getActionDefinitions(G_BRINDIRECT).legalFor({p0}).unsupported(); // Select - setLegalizeScalarToDifferentSizeStrategy(G_SELECT, 0, widen_1_8_16); - - for (auto Ty : {s32, s64, p0}) - setAction({G_SELECT, Ty}, LegalizeAction::Legal); - - setAction({G_SELECT, 1, s1}, LegalizeAction::Legal); + getActionDefinitions(G_SELECT) + .legalFor({{s32, s1}, {s64, s1}, {p0, s1}}) + .clampScalar(0, s32, s64) + .widenScalarToPow2(0) + .unsupported(); // Pointer-handling - setAction({G_FRAME_INDEX, p0}, LegalizeAction::Legal); - setAction({G_GLOBAL_VALUE, p0}, LegalizeAction::Legal); - - for (auto Ty : {s1, s8, s16, s32, s64}) - setAction({G_PTRTOINT, 0, Ty}, LegalizeAction::Legal); - - setAction({G_PTRTOINT, 1, p0}, LegalizeAction::Legal); - - setAction({G_INTTOPTR, 0, p0}, LegalizeAction::Legal); - setAction({G_INTTOPTR, 1, s64}, LegalizeAction::Legal); + getActionDefinitions(G_FRAME_INDEX).legalFor({p0}).unsupported(); + getActionDefinitions(G_GLOBAL_VALUE).legalFor({p0}).unsupported(); + + getActionDefinitions(G_PTRTOINT) + .legalForCartesianProd({s1, s8, s16, s32, s64}, {p0}) + .maxScalar(0, s64) + .widenScalarToPow2(0, /*Min*/ 8) + .unsupported(); + + getActionDefinitions(G_INTTOPTR) + .unsupportedIf([&](const LegalityQuery &Query) { + return Query.Types[0].getSizeInBits() != Query.Types[1].getSizeInBits(); + }) + .legalFor({s64, p0}) + .unsupported(); // Casts for 32 and 64-bit width type are just copies. // Same for 128-bit width type, except they are on the FPR bank. - for (auto Ty : {s1, s8, s16, s32, s64, s128}) { - setAction({G_BITCAST, 0, Ty}, LegalizeAction::Legal); - setAction({G_BITCAST, 1, Ty}, LegalizeAction::Legal); - } - - // For the sake of copying bits around, the type does not really - // matter as long as it fits a register. - for (int EltSize = 8; EltSize <= 64; EltSize *= 2) { - setAction({G_BITCAST, 0, LLT::vector(128/EltSize, EltSize)}, LegalizeAction::Legal); - setAction({G_BITCAST, 1, LLT::vector(128/EltSize, EltSize)}, LegalizeAction::Legal); - if (EltSize >= 64) - continue; - - setAction({G_BITCAST, 0, LLT::vector(64/EltSize, EltSize)}, LegalizeAction::Legal); - setAction({G_BITCAST, 1, LLT::vector(64/EltSize, EltSize)}, LegalizeAction::Legal); - if (EltSize >= 32) - continue; - - setAction({G_BITCAST, 0, LLT::vector(32/EltSize, EltSize)}, LegalizeAction::Legal); - setAction({G_BITCAST, 1, LLT::vector(32/EltSize, EltSize)}, LegalizeAction::Legal); - } + getActionDefinitions(G_BITCAST) + // FIXME: This is wrong since G_BITCAST is not allowed to change the + // number of bits but it's what the previous code described and fixing + // it breaks tests. + .legalForCartesianProd({s1, s8, s16, s32, s64, s128, v16s8, v8s8, v4s8, + v8s16, v4s16, v2s16, v4s32, v2s32, v2s64}) + .unsupported(); - setAction({G_VASTART, p0}, LegalizeAction::Legal); + getActionDefinitions(G_VASTART).legalFor({p0}).unsupported(); // va_list must be a pointer, but most sized types are pretty easy to handle // as the destination. - setAction({G_VAARG, 1, p0}, LegalizeAction::Legal); + getActionDefinitions(G_VAARG) + .customForCartesianProd({s8, s16, s32, s64, p0}, {p0}) + .clampScalar(0, s8, s64) + .widenScalarToPow2(0, /*Min*/ 8) + .unsupported(); - for (auto Ty : {s8, s16, s32, s64, p0}) - setAction({G_VAARG, Ty}, LegalizeAction::Custom); + if (ST.hasLSE()) { + getActionDefinitions(G_ATOMIC_CMPXCHG) + .legalForCartesianProd({s8, s16, s32, s64}, {p0}); + } + getActionDefinitions(G_ATOMIC_CMPXCHG) + .unsupported(); if (ST.hasLSE()) { for (auto Ty : {s8, s16, s32, s64}) { setAction({G_ATOMIC_CMPXCHG_WITH_SUCCESS, Ty}, LegalizeAction::Lower); - setAction({G_ATOMIC_CMPXCHG, Ty}, LegalizeAction::Legal); } - setAction({G_ATOMIC_CMPXCHG, 1, p0}, LegalizeAction::Legal); for (unsigned Op : {G_ATOMICRMW_XCHG, G_ATOMICRMW_ADD, G_ATOMICRMW_SUB, G_ATOMICRMW_AND, G_ATOMICRMW_OR, G_ATOMICRMW_XOR, G_ATOMICRMW_MIN, G_ATOMICRMW_MAX, G_ATOMICRMW_UMIN, G_ATOMICRMW_UMAX}) { - for (auto Ty : {s8, s16, s32, s64}) { - setAction({Op, Ty}, LegalizeAction::Legal); - } - setAction({Op, 1, p0}, LegalizeAction::Legal); + getActionDefinitions(Op).legalForCartesianProd({s8, s16, s32, s64}, {p0}); } } + for (unsigned Op : + {G_ATOMICRMW_XCHG, G_ATOMICRMW_ADD, G_ATOMICRMW_SUB, G_ATOMICRMW_AND, + G_ATOMICRMW_OR, G_ATOMICRMW_XOR, G_ATOMICRMW_MIN, G_ATOMICRMW_MAX, + G_ATOMICRMW_UMIN, G_ATOMICRMW_UMAX}) { + getActionDefinitions(Op).unsupported(); + } // Merge/Unmerge - for (unsigned Op : {G_MERGE_VALUES, G_UNMERGE_VALUES}) - for (int Sz : {8, 16, 32, 64, 128, 192, 256, 384, 512}) { - LLT ScalarTy = LLT::scalar(Sz); - setAction({Op, ScalarTy}, LegalizeAction::Legal); - setAction({Op, 1, ScalarTy}, LegalizeAction::Legal); - if (Sz < 32) - continue; - for (int EltSize = 8; EltSize <= 64; EltSize *= 2) { - if (EltSize >= Sz) - continue; - LLT VecTy = LLT::vector(Sz / EltSize, EltSize); - setAction({Op, VecTy}, LegalizeAction::Legal); - setAction({Op, 1, VecTy}, LegalizeAction::Legal); + for (unsigned Op : {G_MERGE_VALUES, G_UNMERGE_VALUES}) { + unsigned BigTyIdx = Op == G_MERGE_VALUES ? 0 : 1; + unsigned LitTyIdx = Op == G_MERGE_VALUES ? 1 : 0; + + auto notValidElt = [](const LegalityQuery &Query, unsigned TypeIdx) { + const LLT &Ty = Query.Types[TypeIdx]; + if (Ty.isVector()) { + const LLT &EltTy = Ty.getElementType(); + if (EltTy.getSizeInBits() < 8 || EltTy.getSizeInBits() > 64) + return true; + if (!isPowerOf2_32(EltTy.getSizeInBits())) + return true; } - } + return false; + }; + auto scalarize = + [](const LegalityQuery &Query, unsigned TypeIdx) { + const LLT &Ty = Query.Types[TypeIdx]; + return std::make_pair(TypeIdx, Ty.getElementType()); + }; + + // FIXME: This rule is horrible, but specifies the same as what we had + // before with the particularly strange definitions removed (e.g. + // s8 = G_MERGE_VALUES s32, s32). + // Part of the complexity comes from these ops being extremely flexible. For + // example, you can build/decompose vectors with it, concatenate vectors, + // etc. and in addition to this you can also bitcast with it at the same + // time. We've been considering breaking it up into multiple ops to make it + // more manageable throughout the backend. + getActionDefinitions(Op) + // Break up vectors with weird elements into scalars + .fewerElementsIf( + [=](const LegalityQuery &Query) { return notValidElt(Query, 0); }, + [=](const LegalityQuery &Query) { return scalarize(Query, 0); }) + .fewerElementsIf( + [=](const LegalityQuery &Query) { return notValidElt(Query, 1); }, + [=](const LegalityQuery &Query) { return scalarize(Query, 1); }) + // Clamp the big scalar to s8-s512 and make it either a power of 2, 192, + // or 384. + .clampScalar(BigTyIdx, s8, s512) + .widenScalarIf( + [=](const LegalityQuery &Query) { + const LLT &Ty = Query.Types[BigTyIdx]; + return !isPowerOf2_32(Ty.getSizeInBits()) && + Ty.getSizeInBits() % 64 != 0; + }, + [=](const LegalityQuery &Query) { + // Pick the next power of 2, or a multiple of 64 over 128. + // Whichever is smaller. + const LLT &Ty = Query.Types[BigTyIdx]; + unsigned NewSizeInBits = 1 + << Log2_32_Ceil(Ty.getSizeInBits() + 1); + if (NewSizeInBits >= 256) { + unsigned RoundedTo = alignTo<64>(Ty.getSizeInBits() + 1); + if (RoundedTo < NewSizeInBits) + NewSizeInBits = RoundedTo; + } + return std::make_pair(BigTyIdx, LLT::scalar(NewSizeInBits)); + }) + // Clamp the little scalar to s8-s256 and make it a power of 2. It's not + // worth considering the multiples of 64 since 2*192 and 2*384 are not + // valid. + .clampScalar(LitTyIdx, s8, s256) + .widenScalarToPow2(LitTyIdx, /*Min*/ 8) + // So at this point, we have s8, s16, s32, s64, s128, s192, s256, s384, + // s512, , , , or . + // At this point it's simple enough to accept the legal types. + .legalIf([=](const LegalityQuery &Query) { + const LLT &BigTy = Query.Types[BigTyIdx]; + const LLT &LitTy = Query.Types[LitTyIdx]; + if (BigTy.isVector() && BigTy.getSizeInBits() < 32) + return false; + if (LitTy.isVector() && LitTy.getSizeInBits() < 32) + return false; + return BigTy.getSizeInBits() % LitTy.getSizeInBits() == 0; + }) + // Any vectors left are the wrong size. Scalarize them. + .fewerElementsIf([](const LegalityQuery &Query) { return true; }, + [](const LegalityQuery &Query) { + return std::make_pair( + 0, Query.Types[0].getElementType()); + }) + .fewerElementsIf([](const LegalityQuery &Query) { return true; }, + [](const LegalityQuery &Query) { + return std::make_pair( + 1, Query.Types[1].getElementType()); + }) + .unsupported(); + } computeTables(); }