Files
swift-mirror/lib/SIL/IR/SILBuilder.cpp
Michael Gottesman e1a19e4173 [sil] Split library into subfolders, while still building as a single library still.
Specifically, I split it into 3 initial categories: IR, Utils, Verifier. I just
did this quickly, we can always split it more later if we want.

I followed the model that we use in SILOptimizer: ./lib/SIL/CMakeLists.txt vends
 a macro (sil_register_sources) to the sub-folders that register the sources of
 the subdirectory with a global state variable that ./lib/SIL/CMakeLists.txt
 defines. Then after including those subdirs, the parent cmake declares the SIL
 library. So the output is the same, but we have the flexibility of having
 subdirectories to categorize source files.
2020-03-30 11:01:00 -07:00

640 lines
25 KiB
C++

//===--- SILBuilder.cpp - Class for creating SIL Constructs ---------------===//
//
// This source file is part of the Swift.org open source project
//
// Copyright (c) 2014 - 2017 Apple Inc. and the Swift project authors
// Licensed under Apache License v2.0 with Runtime Library Exception
//
// See https://swift.org/LICENSE.txt for license information
// See https://swift.org/CONTRIBUTORS.txt for the list of Swift project authors
//
//===----------------------------------------------------------------------===//
#include "swift/SIL/SILBuilder.h"
#include "swift/AST/Expr.h"
#include "swift/SIL/Projection.h"
#include "swift/SIL/SILGlobalVariable.h"
using namespace swift;
//===----------------------------------------------------------------------===//
// SILBuilder Implementation
//===----------------------------------------------------------------------===//
SILBuilder::SILBuilder(SILGlobalVariable *GlobVar,
SmallVectorImpl<SILInstruction *> *InsertedInstrs)
: TempContext(GlobVar->getModule(), InsertedInstrs), C(TempContext),
F(nullptr) {
setInsertionPoint(&GlobVar->StaticInitializerBlock);
}
IntegerLiteralInst *SILBuilder::createIntegerLiteral(IntegerLiteralExpr *E) {
return insert(IntegerLiteralInst::create(E, getSILDebugLocation(E),
getModule()));
}
FloatLiteralInst *SILBuilder::createFloatLiteral(FloatLiteralExpr *E) {
return insert(FloatLiteralInst::create(E, getSILDebugLocation(E),
getModule()));
}
TupleInst *SILBuilder::createTuple(SILLocation loc, ArrayRef<SILValue> elts) {
// Derive the tuple type from the elements.
SmallVector<TupleTypeElt, 4> eltTypes;
for (auto elt : elts)
eltTypes.push_back(elt->getType().getASTType());
auto tupleType = SILType::getPrimitiveObjectType(
CanType(TupleType::get(eltTypes, getASTContext())));
return createTuple(loc, tupleType, elts);
}
SILType SILBuilder::getPartialApplyResultType(
TypeExpansionContext context, SILType origTy, unsigned argCount,
SILModule &M, SubstitutionMap subs, ParameterConvention calleeConvention,
PartialApplyInst::OnStackKind onStack) {
CanSILFunctionType FTI = origTy.castTo<SILFunctionType>();
if (!subs.empty())
FTI = FTI->substGenericArgs(M, subs, context);
assert(!FTI->isPolymorphic()
&& "must provide substitutions for generic partial_apply");
auto params = FTI->getParameters();
auto newParams = params.slice(0, params.size() - argCount);
auto extInfo = FTI->getExtInfo()
.withRepresentation(SILFunctionType::Representation::Thick)
.withIsPseudogeneric(false);
if (onStack)
extInfo = extInfo.withNoEscape();
// If the original method has an @unowned_inner_pointer return, the partial
// application thunk will lifetime-extend 'self' for us, converting the
// return value to @unowned.
//
// If the original method has an @autoreleased return, the partial application
// thunk will retain it for us, converting the return value to @owned.
SmallVector<SILResultInfo, 4> results;
results.append(FTI->getResults().begin(), FTI->getResults().end());
for (auto &result : results) {
if (result.getConvention() == ResultConvention::UnownedInnerPointer)
result = SILResultInfo(result.getReturnValueType(M, FTI),
ResultConvention::Unowned);
else if (result.getConvention() == ResultConvention::Autoreleased)
result = SILResultInfo(result.getReturnValueType(M, FTI),
ResultConvention::Owned);
}
// Do we still need the substitutions in the result?
bool needsSubstFunctionType = false;
for (auto param : newParams) {
needsSubstFunctionType |= param.getInterfaceType()->hasTypeParameter();
}
for (auto result : results) {
needsSubstFunctionType |= result.getInterfaceType()->hasTypeParameter();
}
for (auto yield : FTI->getYields()) {
needsSubstFunctionType |= yield.getInterfaceType()->hasTypeParameter();
}
SubstitutionMap appliedSubs;
if (needsSubstFunctionType) {
appliedSubs = FTI->getCombinedSubstitutions();
}
auto appliedFnType = SILFunctionType::get(nullptr,
extInfo,
FTI->getCoroutineKind(),
calleeConvention,
newParams,
FTI->getYields(),
results,
FTI->getOptionalErrorResult(),
appliedSubs,
SubstitutionMap(),
M.getASTContext());
return SILType::getPrimitiveObjectType(appliedFnType);
}
ProjectBoxInst *SILBuilder::createProjectBox(SILLocation Loc,
SILValue boxOperand,
unsigned index) {
auto boxTy = boxOperand->getType().castTo<SILBoxType>();
auto fieldTy = getSILBoxFieldType(getTypeExpansionContext(), boxTy,
getModule().Types, index);
return insert(new (getModule()) ProjectBoxInst(
getSILDebugLocation(Loc), boxOperand, index, fieldTy));
}
ClassifyBridgeObjectInst *
SILBuilder::createClassifyBridgeObject(SILLocation Loc, SILValue value) {
auto &ctx = getASTContext();
Type int1Ty = BuiltinIntegerType::get(1, ctx);
Type resultTy = TupleType::get({ int1Ty, int1Ty }, ctx);
auto ty = SILType::getPrimitiveObjectType(resultTy->getCanonicalType());
return insert(new (getModule())
ClassifyBridgeObjectInst(getSILDebugLocation(Loc), value, ty));
}
// Create the appropriate cast instruction based on result type.
SingleValueInstruction *
SILBuilder::createUncheckedBitCast(SILLocation Loc, SILValue Op, SILType Ty) {
assert(isLoadableOrOpaque(Ty));
if (Ty.isTrivial(getFunction()))
return insert(UncheckedTrivialBitCastInst::create(
getSILDebugLocation(Loc), Op, Ty, getFunction(), C.OpenedArchetypes));
if (SILType::canRefCast(Op->getType(), Ty, getModule()))
return createUncheckedRefCast(Loc, Op, Ty);
// The destination type is nontrivial, and may be smaller than the source
// type, so RC identity cannot be assumed.
return insert(UncheckedBitwiseCastInst::create(
getSILDebugLocation(Loc), Op, Ty, getFunction(), C.OpenedArchetypes));
}
BranchInst *SILBuilder::createBranch(SILLocation Loc,
SILBasicBlock *TargetBlock,
OperandValueArrayRef Args) {
SmallVector<SILValue, 6> ArgsCopy;
ArgsCopy.reserve(Args.size());
for (auto I = Args.begin(), E = Args.end(); I != E; ++I)
ArgsCopy.push_back(*I);
return createBranch(Loc, TargetBlock, ArgsCopy);
}
/// Branch to the given block if there's an active insertion point,
/// then move the insertion point to the end of that block.
void SILBuilder::emitBlock(SILBasicBlock *BB, SILLocation BranchLoc) {
if (!hasValidInsertionPoint()) {
return emitBlock(BB);
}
// Fall though from the currently active block into the given block.
assert(BB->args_empty() && "cannot fall through to bb with args");
// This is a fall through into BB, emit the fall through branch.
createBranch(BranchLoc, BB);
// Start inserting into that block.
setInsertionPoint(BB);
}
/// splitBlockForFallthrough - Prepare for the insertion of a terminator. If
/// the builder's insertion point is at the end of the current block (as when
/// SILGen is creating the initial code for a function), just create and
/// return a new basic block that will be later used for the continue point.
///
/// If the insertion point is valid (i.e., pointing to an existing
/// instruction) then split the block at that instruction and return the
/// continuation block.
SILBasicBlock *SILBuilder::splitBlockForFallthrough() {
// If we are concatenating, just create and return a new block.
if (insertingAtEndOfBlock()) {
return getFunction().createBasicBlockAfter(BB);
}
// Otherwise we need to split the current block at the insertion point.
auto *NewBB = BB->split(InsertPt);
InsertPt = BB->end();
return NewBB;
}
static bool setAccessToDeinit(BeginAccessInst *beginAccess) {
// It's possible that AllocBoxToStack could catch some cases that
// AccessEnforcementSelection does not promote to [static]. Ultimately, this
// should be an assert, but only after we the two passes can be fixed to share
// a common analysis.
if (beginAccess->getEnforcement() == SILAccessEnforcement::Dynamic)
return false;
beginAccess->setAccessKind(SILAccessKind::Deinit);
return true;
}
PointerUnion<CopyAddrInst *, DestroyAddrInst *>
SILBuilder::emitDestroyAddr(SILLocation Loc, SILValue Operand) {
// Check to see if the instruction immediately before the insertion point is a
// copy_addr from the specified operand. If so, we can fold this into the
// copy_addr as a take.
BeginAccessInst *beginAccess = nullptr;
CopyAddrInst *copyAddrTake = nullptr;
auto I = getInsertionPoint(), BBStart = getInsertionBB()->begin();
while (I != BBStart) {
auto *Inst = &*--I;
if (auto CA = dyn_cast<CopyAddrInst>(Inst)) {
if (!CA->isTakeOfSrc()) {
if (CA->getSrc() == Operand && !CA->isTakeOfSrc()) {
CA->setIsTakeOfSrc(IsTake);
return CA;
}
// If this copy_addr is accessing the same source, continue searching
// backward until we see the begin_access. If any side effects occur
// between the `%adr = begin_access %src` and `copy_addr %adr` then we
// cannot promote the access to a deinit. `[deinit]` requires exclusive
// access, but an instruction with side effects may require shared
// access.
if (CA->getSrc() == beginAccess) {
copyAddrTake = CA;
continue;
}
}
}
// If we've already seen a copy_addr that can be convert to `take`, then
// stop at the begin_access for the copy's source.
if (copyAddrTake && beginAccess == Inst) {
// If `setAccessToDeinit()` returns `true` it has modified the access
// instruction, so we are committed to the transformation on that path.
if (setAccessToDeinit(beginAccess)) {
copyAddrTake->setIsTakeOfSrc(IsTake);
return copyAddrTake;
}
}
// destroy_addrs commonly exist in a block of dealloc_stack's, which don't
// affect take-ability.
if (isa<DeallocStackInst>(Inst))
continue;
// An end_access of the same address may be able to be rewritten as a
// [deinit] access.
if (auto endAccess = dyn_cast<EndAccessInst>(Inst)) {
if (endAccess->getSource() == Operand) {
beginAccess = endAccess->getBeginAccess();
continue;
}
}
// This code doesn't try to prove tricky validity constraints about whether
// it is safe to push the destroy_addr past interesting instructions.
if (Inst->mayHaveSideEffects())
break;
}
// If we didn't find a copy_addr to fold this into, emit the destroy_addr.
return createDestroyAddr(Loc, Operand);
}
static bool couldReduceStrongRefcount(SILInstruction *Inst) {
// Simple memory accesses cannot reduce refcounts.
switch (Inst->getKind()) {
#define NEVER_LOADABLE_CHECKED_REF_STORAGE(Name, ...) \
case SILInstructionKind::Store##Name##Inst: \
return false;
#define ALWAYS_LOADABLE_CHECKED_REF_STORAGE(Name, ...) \
/* The next case must be first in this macro because */ \
/* SOMETIMES_LOADABLE_CHECKED_REF_STORAGE will fall into it. */ \
case SILInstructionKind::Name##ReleaseInst: \
if (isLessStrongThan(ReferenceOwnership::Name, ReferenceOwnership::Strong))\
return false; \
break; \
case SILInstructionKind::Name##RetainInst: \
case SILInstructionKind::StrongRetain##Name##Inst: \
return false;
#define SOMETIMES_LOADABLE_CHECKED_REF_STORAGE(Name, ...) \
case SILInstructionKind::Store##Name##Inst: \
ALWAYS_LOADABLE_CHECKED_REF_STORAGE(Name, "...")
#define UNCHECKED_REF_STORAGE(Name, ...) \
case SILInstructionKind::StrongCopy##Name##ValueInst: \
return false;
#include "swift/AST/ReferenceStorage.def"
case SILInstructionKind::LoadInst:
case SILInstructionKind::StoreInst:
case SILInstructionKind::RetainValueInst:
case SILInstructionKind::StrongRetainInst:
case SILInstructionKind::AllocStackInst:
case SILInstructionKind::DeallocStackInst:
return false;
default:
break;
}
// Assign and copyaddr of trivial types cannot drop refcounts, and 'inits'
// never can either. Nontrivial ones can though, because the overwritten
// value drops a retain. We would have to do more alias analysis to be able
// to safely ignore one of those.
if (auto AI = dyn_cast<AssignInst>(Inst)) {
auto StoredType = AI->getOperand(0)->getType();
if (StoredType.isTrivial(*Inst->getFunction()) ||
StoredType.is<ReferenceStorageType>())
return false;
}
if (auto *CAI = dyn_cast<CopyAddrInst>(Inst)) {
// Initializations can only increase refcounts.
if (CAI->isInitializationOfDest())
return false;
SILType StoredType = CAI->getOperand(0)->getType().getObjectType();
if (StoredType.isTrivial(*Inst->getFunction()) ||
StoredType.is<ReferenceStorageType>())
return false;
}
// This code doesn't try to prove tricky validity constraints about whether
// it is safe to push the release past interesting instructions.
return Inst->mayHaveSideEffects();
}
/// Perform a strong_release instruction at the current location, attempting
/// to fold it locally into nearby retain instructions or emitting an explicit
/// strong release if necessary. If this inserts a new instruction, it
/// returns it, otherwise it returns null.
PointerUnion<StrongRetainInst *, StrongReleaseInst *>
SILBuilder::emitStrongRelease(SILLocation Loc, SILValue Operand) {
// Release on a functionref is a noop.
if (isa<FunctionRefInst>(Operand)) {
return static_cast<StrongReleaseInst *>(nullptr);
}
// Check to see if the instruction immediately before the insertion point is a
// strong_retain of the specified operand. If so, we can zap the pair.
auto I = getInsertionPoint(), BBStart = getInsertionBB()->begin();
while (I != BBStart) {
auto *Inst = &*--I;
if (auto *SRA = dyn_cast<StrongRetainInst>(Inst)) {
if (SRA->getOperand() == Operand)
return SRA;
// Skip past unrelated retains.
continue;
}
// Scan past simple instructions that cannot reduce strong refcounts.
if (couldReduceStrongRefcount(Inst))
break;
}
// If we didn't find a retain to fold this into, emit the release.
return createStrongRelease(Loc, Operand, getDefaultAtomicity());
}
/// Emit a release_value instruction at the current location, attempting to
/// fold it locally into another nearby retain_value instruction. This
/// returns the new instruction if it inserts one, otherwise it returns null.
PointerUnion<RetainValueInst *, ReleaseValueInst *>
SILBuilder::emitReleaseValue(SILLocation Loc, SILValue Operand) {
// Check to see if the instruction immediately before the insertion point is a
// retain_value of the specified operand. If so, we can zap the pair.
auto I = getInsertionPoint(), BBStart = getInsertionBB()->begin();
while (I != BBStart) {
auto *Inst = &*--I;
if (auto *SRA = dyn_cast<RetainValueInst>(Inst)) {
if (SRA->getOperand() == Operand)
return SRA;
// Skip past unrelated retains.
continue;
}
// Scan past simple instructions that cannot reduce refcounts.
if (couldReduceStrongRefcount(Inst))
break;
}
// If we didn't find a retain to fold this into, emit the release.
return createReleaseValue(Loc, Operand, getDefaultAtomicity());
}
PointerUnion<CopyValueInst *, DestroyValueInst *>
SILBuilder::emitDestroyValue(SILLocation Loc, SILValue Operand) {
// Check to see if the instruction immediately before the insertion point is a
// retain_value of the specified operand. If so, we can zap the pair.
auto I = getInsertionPoint(), BBStart = getInsertionBB()->begin();
while (I != BBStart) {
auto *Inst = &*--I;
if (auto *CVI = dyn_cast<CopyValueInst>(Inst)) {
if (SILValue(CVI) == Operand || CVI->getOperand() == Operand)
return CVI;
// Skip past unrelated retains.
continue;
}
// Scan past simple instructions that cannot reduce refcounts.
if (couldReduceStrongRefcount(Inst))
break;
}
// If we didn't find a retain to fold this into, emit the release.
return createDestroyValue(Loc, Operand);
}
SILValue SILBuilder::emitThickToObjCMetatype(SILLocation Loc, SILValue Op,
SILType Ty) {
// If the operand is an otherwise-unused 'metatype' instruction in the
// same basic block, zap it and create a 'metatype' instruction that
// directly produces an Objective-C metatype.
if (auto metatypeInst = dyn_cast<MetatypeInst>(Op)) {
if (metatypeInst->use_empty() &&
metatypeInst->getParent() == getInsertionBB()) {
auto origLoc = metatypeInst->getLoc();
metatypeInst->eraseFromParent();
return createMetatype(origLoc, Ty);
}
}
// Just create the thick_to_objc_metatype instruction.
return createThickToObjCMetatype(Loc, Op, Ty);
}
SILValue SILBuilder::emitObjCToThickMetatype(SILLocation Loc, SILValue Op,
SILType Ty) {
// If the operand is an otherwise-unused 'metatype' instruction in the
// same basic block, zap it and create a 'metatype' instruction that
// directly produces a thick metatype.
if (auto metatypeInst = dyn_cast<MetatypeInst>(Op)) {
if (metatypeInst->use_empty() &&
metatypeInst->getParent() == getInsertionBB()) {
auto origLoc = metatypeInst->getLoc();
metatypeInst->eraseFromParent();
return createMetatype(origLoc, Ty);
}
}
// Just create the objc_to_thick_metatype instruction.
return createObjCToThickMetatype(Loc, Op, Ty);
}
/// Add opened archetypes defined or used by the current instruction.
/// If there are no such opened archetypes in the current instruction
/// and it is an instruction with just one operand, try to perform
/// the same action for the instruction defining an operand, because
/// it may have some opened archetypes used or defined.
void SILBuilder::addOpenedArchetypeOperands(SILInstruction *I) {
// The list of archetypes from the previous instruction needs
// to be replaced, because it may reference a removed instruction.
C.OpenedArchetypes.addOpenedArchetypeOperands(I->getTypeDependentOperands());
if (I && I->getNumTypeDependentOperands() > 0)
return;
// Keep track of already visited instructions to avoid infinite loops.
SmallPtrSet<SILInstruction *, 8> Visited;
while (I && I->getNumOperands() == 1 &&
I->getNumTypeDependentOperands() == 0) {
// All the open instructions are single-value instructions. Operands may
// be null when code is being transformed.
auto SVI = dyn_cast_or_null<SingleValueInstruction>(I->getOperand(0));
// Within SimplifyCFG this function may be called for an instruction
// within unreachable code. And within an unreachable block it can happen
// that defs do not dominate uses (because there is no dominance defined).
// To avoid the infinite loop when following the chain of instructions via
// their operands, bail if the operand is not an instruction or this
// instruction was seen already.
if (!SVI || !Visited.insert(SVI).second)
return;
// If it is a definition of an opened archetype,
// register it and exit.
auto Archetype = getOpenedArchetypeOf(SVI);
if (!Archetype) {
I = SVI;
continue;
}
auto Def = C.OpenedArchetypes.getOpenedArchetypeDef(Archetype);
// Return if it is a known open archetype.
if (Def)
return;
// Otherwise register it and return.
if (C.OpenedArchetypesTracker)
C.OpenedArchetypesTracker->addOpenedArchetypeDef(Archetype, SVI);
return;
}
if (I && I->getNumTypeDependentOperands() > 0) {
C.OpenedArchetypes.addOpenedArchetypeOperands(
I->getTypeDependentOperands());
}
}
ValueMetatypeInst *SILBuilder::createValueMetatype(SILLocation Loc,
SILType MetatypeTy,
SILValue Base) {
assert(Base->getType().isLoweringOf(
getTypeExpansionContext(), getModule(),
MetatypeTy.castTo<MetatypeType>().getInstanceType()) &&
"value_metatype result must be formal metatype of the lowered operand "
"type");
return insert(new (getModule()) ValueMetatypeInst(getSILDebugLocation(Loc),
MetatypeTy, Base));
}
// TODO: This should really be an operation on type lowering.
void SILBuilder::emitDestructureValueOperation(
SILLocation loc, SILValue v, SmallVectorImpl<SILValue> &results) {
// Once destructure is allowed everywhere, remove the projection code.
// If we do not have a tuple or a struct, add to our results list and return.
SILType type = v->getType();
if (!(type.is<TupleType>() || type.getStructOrBoundGenericStruct())) {
results.emplace_back(v);
return;
}
// Otherwise, we want to destructure add the destructure and return.
if (getFunction().hasOwnership()) {
auto *i = emitDestructureValueOperation(loc, v);
llvm::copy(i->getResults(), std::back_inserter(results));
return;
}
// In non qualified ownership SIL, drop back to using projection code.
SmallVector<Projection, 16> projections;
Projection::getFirstLevelProjections(v->getType(), getModule(),
getTypeExpansionContext(), projections);
llvm::transform(projections, std::back_inserter(results),
[&](const Projection &p) -> SILValue {
return p.createObjectProjection(*this, loc, v).get();
});
}
// TODO: Can we put this on type lowering? It would take a little bit of work
// since we would need to be able to handle aggregate trivial types which is not
// represented today in TypeLowering.
void SILBuilder::emitDestructureAddressOperation(
SILLocation loc, SILValue v, SmallVectorImpl<SILValue> &results) {
// If we do not have a tuple or a struct, add to our results list.
SILType type = v->getType();
if (!(type.is<TupleType>() || type.getStructOrBoundGenericStruct())) {
results.emplace_back(v);
return;
}
SmallVector<Projection, 16> projections;
Projection::getFirstLevelProjections(v->getType(), getModule(),
getTypeExpansionContext(), projections);
llvm::transform(projections, std::back_inserter(results),
[&](const Projection &p) -> SILValue {
return p.createAddressProjection(*this, loc, v).get();
});
}
void SILBuilder::emitDestructureValueOperation(
SILLocation loc, SILValue operand,
function_ref<void(unsigned, SILValue)> func) {
// Do a quick check to see if we have a tuple without elements. In that
// case, bail early since we are not going to ever invoke Func.
if (auto tupleType = operand->getType().getAs<TupleType>())
if (0 == tupleType->getNumElements())
return;
SmallVector<SILValue, 8> results;
emitDestructureValueOperation(loc, operand, results);
for (auto p : llvm::enumerate(results)) {
func(p.index(), p.value());
}
}
DebugValueInst *SILBuilder::createDebugValue(SILLocation Loc, SILValue src,
SILDebugVariable Var) {
assert(isLoadableOrOpaque(src->getType()));
// Debug location overrides cannot apply to debug value instructions.
DebugLocOverrideRAII LocOverride{*this, None};
return insert(
DebugValueInst::create(getSILDebugLocation(Loc), src, getModule(), Var));
}
DebugValueAddrInst *SILBuilder::createDebugValueAddr(SILLocation Loc,
SILValue src,
SILDebugVariable Var) {
// Debug location overrides cannot apply to debug addr instructions.
DebugLocOverrideRAII LocOverride{*this, None};
return insert(DebugValueAddrInst::create(getSILDebugLocation(Loc), src,
getModule(), Var));
}
void SILBuilder::emitScopedBorrowOperation(SILLocation loc, SILValue original,
function_ref<void(SILValue)> &&fun) {
if (original->getType().isAddress()) {
original = createLoadBorrow(loc, original);
} else {
original = createBeginBorrow(loc, original);
}
fun(original);
createEndBorrow(loc, original);
}
CheckedCastBranchInst *SILBuilder::createCheckedCastBranch(
SILLocation Loc, bool isExact, SILValue op,
SILType destLoweredTy, CanType destFormalTy,
SILBasicBlock *successBB, SILBasicBlock *failureBB,
ProfileCounter target1Count, ProfileCounter target2Count) {
assert((!hasOwnership() || !failureBB->getNumArguments() ||
failureBB->getArgument(0)->getType() == op->getType()) &&
"failureBB's argument doesn't match incoming argument type");
return insertTerminator(CheckedCastBranchInst::create(
getSILDebugLocation(Loc), isExact, op,
destLoweredTy, destFormalTy, successBB, failureBB,
getFunction(), C.OpenedArchetypes, target1Count, target2Count));
}