ComplexPairTy ComplexExprEmitter::
VisitAbstractConditionalOperator(const AbstractConditionalOperator *E) {
TestAndClearIgnoreReal();
TestAndClearIgnoreImag();
llvm::BasicBlock *LHSBlock = CGF.createBasicBlock("cond.true");
llvm::BasicBlock *RHSBlock = CGF.createBasicBlock("cond.false");
llvm::BasicBlock *ContBlock = CGF.createBasicBlock("cond.end");
// Bind the common expression if necessary.
CodeGenFunction::OpaqueValueMapping binding(CGF, E);
RegionCounter Cnt = CGF.getPGORegionCounter(E);
CodeGenFunction::ConditionalEvaluation eval(CGF);
CGF.EmitBranchOnBoolExpr(E->getCond(), LHSBlock, RHSBlock, Cnt.getCount());
eval.begin(CGF);
CGF.EmitBlock(LHSBlock);
Cnt.beginRegion(Builder);
ComplexPairTy LHS = Visit(E->getTrueExpr());
LHSBlock = Builder.GetInsertBlock();
CGF.EmitBranch(ContBlock);
eval.end(CGF);
eval.begin(CGF);
CGF.EmitBlock(RHSBlock);
ComplexPairTy RHS = Visit(E->getFalseExpr());
RHSBlock = Builder.GetInsertBlock();
CGF.EmitBlock(ContBlock);
eval.end(CGF);
// Create a PHI node for the real part.
llvm::PHINode *RealPN = Builder.CreatePHI(LHS.first->getType(), 2, "cond.r");
RealPN->addIncoming(LHS.first, LHSBlock);
RealPN->addIncoming(RHS.first, RHSBlock);
// Create a PHI node for the imaginary part.
llvm::PHINode *ImagPN = Builder.CreatePHI(LHS.first->getType(), 2, "cond.i");
ImagPN->addIncoming(LHS.second, LHSBlock);
ImagPN->addIncoming(RHS.second, RHSBlock);
return ComplexPairTy(RealPN, ImagPN);
}
void CodeGenFunction::EmitOMPSimdDirective(const OMPSimdDirective &S) {
// Pragma 'simd' code depends on presence of 'lastprivate'.
// If present, we have to separate last iteration of the loop:
//
// if (LastIteration != 0) {
// for (IV in 0..LastIteration-1) BODY;
// BODY with updates of lastprivate vars;
// <Final counter/linear vars updates>;
// }
//
// otherwise (when there's no lastprivate):
//
// for (IV in 0..LastIteration) BODY;
// <Final counter/linear vars updates>;
//
// Walk clauses and process safelen/lastprivate.
bool SeparateIter = false;
LoopStack.setParallel();
LoopStack.setVectorizerEnable(true);
for (auto C : S.clauses()) {
switch (C->getClauseKind()) {
case OMPC_safelen: {
RValue Len = EmitAnyExpr(cast<OMPSafelenClause>(C)->getSafelen(),
AggValueSlot::ignored(), true);
llvm::ConstantInt *Val = cast<llvm::ConstantInt>(Len.getScalarVal());
LoopStack.setVectorizerWidth(Val->getZExtValue());
// In presence of finite 'safelen', it may be unsafe to mark all
// the memory instructions parallel, because loop-carried
// dependences of 'safelen' iterations are possible.
LoopStack.setParallel(false);
break;
}
case OMPC_aligned:
EmitOMPAlignedClause(*this, CGM, cast<OMPAlignedClause>(*C));
break;
case OMPC_lastprivate:
SeparateIter = true;
break;
default:
// Not handled yet
;
}
}
RunCleanupsScope DirectiveScope(*this);
CGDebugInfo *DI = getDebugInfo();
if (DI)
DI->EmitLexicalBlockStart(Builder, S.getSourceRange().getBegin());
// Emit the loop iteration variable.
const Expr *IVExpr = S.getIterationVariable();
const VarDecl *IVDecl = cast<VarDecl>(cast<DeclRefExpr>(IVExpr)->getDecl());
EmitVarDecl(*IVDecl);
EmitIgnoredExpr(S.getInit());
// Emit the iterations count variable.
// If it is not a variable, Sema decided to calculate iterations count on each
// iteration (e.g., it is foldable into a constant).
if (auto LIExpr = dyn_cast<DeclRefExpr>(S.getLastIteration())) {
EmitVarDecl(*cast<VarDecl>(LIExpr->getDecl()));
// Emit calculation of the iterations count.
EmitIgnoredExpr(S.getCalcLastIteration());
}
if (SeparateIter) {
// Emit: if (LastIteration > 0) - begin.
RegionCounter Cnt = getPGORegionCounter(&S);
auto ThenBlock = createBasicBlock("simd.if.then");
auto ContBlock = createBasicBlock("simd.if.end");
EmitBranchOnBoolExpr(S.getPreCond(), ThenBlock, ContBlock, Cnt.getCount());
EmitBlock(ThenBlock);
Cnt.beginRegion(Builder);
// Emit 'then' code.
{
OMPPrivateScope LoopScope(*this);
EmitPrivateLoopCounters(*this, LoopScope, S.counters());
EmitOMPInnerLoop(S, LoopScope, /* SeparateIter */ true);
EmitOMPLoopBody(S, /* SeparateIter */ true);
}
EmitOMPSimdFinal(S);
// Emit: if (LastIteration != 0) - end.
EmitBranch(ContBlock);
EmitBlock(ContBlock, true);
} else {
{
OMPPrivateScope LoopScope(*this);
EmitPrivateLoopCounters(*this, LoopScope, S.counters());
EmitOMPInnerLoop(S, LoopScope);
}
EmitOMPSimdFinal(S);
}
if (DI)
DI->EmitLexicalBlockEnd(Builder, S.getSourceRange().getEnd());
}
void CodeGenFunction::EmitOMPWorksharingLoop(const OMPLoopDirective &S) {
// Emit the loop iteration variable.
auto IVExpr = cast<DeclRefExpr>(S.getIterationVariable());
auto IVDecl = cast<VarDecl>(IVExpr->getDecl());
EmitVarDecl(*IVDecl);
// Emit the iterations count variable.
// If it is not a variable, Sema decided to calculate iterations count on each
// iteration (e.g., it is foldable into a constant).
if (auto LIExpr = dyn_cast<DeclRefExpr>(S.getLastIteration())) {
EmitVarDecl(*cast<VarDecl>(LIExpr->getDecl()));
// Emit calculation of the iterations count.
EmitIgnoredExpr(S.getCalcLastIteration());
}
auto &RT = CGM.getOpenMPRuntime();
// Check pre-condition.
{
// Skip the entire loop if we don't meet the precondition.
RegionCounter Cnt = getPGORegionCounter(&S);
auto ThenBlock = createBasicBlock("omp.precond.then");
auto ContBlock = createBasicBlock("omp.precond.end");
EmitBranchOnBoolExpr(S.getPreCond(), ThenBlock, ContBlock, Cnt.getCount());
EmitBlock(ThenBlock);
Cnt.beginRegion(Builder);
// Emit 'then' code.
{
// Emit helper vars inits.
LValue LB =
EmitOMPHelperVar(*this, cast<DeclRefExpr>(S.getLowerBoundVariable()));
LValue UB =
EmitOMPHelperVar(*this, cast<DeclRefExpr>(S.getUpperBoundVariable()));
LValue ST =
EmitOMPHelperVar(*this, cast<DeclRefExpr>(S.getStrideVariable()));
LValue IL =
EmitOMPHelperVar(*this, cast<DeclRefExpr>(S.getIsLastIterVariable()));
OMPPrivateScope LoopScope(*this);
EmitPrivateLoopCounters(*this, LoopScope, S.counters());
// Detect the loop schedule kind and chunk.
auto ScheduleKind = OMPC_SCHEDULE_unknown;
llvm::Value *Chunk = nullptr;
if (auto C = cast_or_null<OMPScheduleClause>(
S.getSingleClause(OMPC_schedule))) {
ScheduleKind = C->getScheduleKind();
if (auto Ch = C->getChunkSize()) {
Chunk = EmitScalarExpr(Ch);
Chunk = EmitScalarConversion(Chunk, Ch->getType(),
S.getIterationVariable()->getType());
}
}
const unsigned IVSize = getContext().getTypeSize(IVExpr->getType());
const bool IVSigned = IVExpr->getType()->hasSignedIntegerRepresentation();
if (RT.isStaticNonchunked(ScheduleKind,
/* Chunked */ Chunk != nullptr)) {
// OpenMP [2.7.1, Loop Construct, Description, table 2-1]
// When no chunk_size is specified, the iteration space is divided into
// chunks that are approximately equal in size, and at most one chunk is
// distributed to each thread. Note that the size of the chunks is
// unspecified in this case.
RT.EmitOMPForInit(*this, S.getLocStart(), ScheduleKind, IVSize, IVSigned,
IL.getAddress(), LB.getAddress(), UB.getAddress(),
ST.getAddress());
// UB = min(UB, GlobalUB);
EmitIgnoredExpr(S.getEnsureUpperBound());
// IV = LB;
EmitIgnoredExpr(S.getInit());
// while (idx <= UB) { BODY; ++idx; }
EmitOMPInnerLoop(S, LoopScope);
// Tell the runtime we are done.
RT.EmitOMPForFinish(*this, S.getLocStart(), ScheduleKind);
} else {
// Emit the outer loop, which requests its work chunk [LB..UB] from
// runtime and runs the inner loop to process it.
EmitOMPForOuterLoop(ScheduleKind, S, LoopScope, LB.getAddress(),
UB.getAddress(), ST.getAddress(), IL.getAddress(),
Chunk);
}
}
// We're now done with the loop, so jump to the continuation block.
EmitBranch(ContBlock);
EmitBlock(ContBlock, true);
}
}
