[llvm-branch-commits] [flang] [Flang][OpenMP][Lower] Split MLIR codegen for clauses and constructs (PR #86963)
Sergio Afonso via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Thu Mar 28 09:53:24 PDT 2024
https://github.com/skatrak updated https://github.com/llvm/llvm-project/pull/86963
>From e291fad68b78d28bfa73caab94ddcb978db2a602 Mon Sep 17 00:00:00 2001
From: Sergio Afonso <safonsof at amd.com>
Date: Thu, 28 Mar 2024 15:14:37 +0000
Subject: [PATCH] [Flang][OpenMP][Lower] Split MLIR codegen for clauses and
constructs
This patch performs several cleanups with the main purpose of normalizing the
code patterns used to trigger codegen for MLIR OpenMP operations and making the
processing of clauses and constructs independent. The following changes are
made:
- Clean up unused `directive` argument to `ClauseProcessor::processMap()`.
- Move general helper functions in OpenMP.cpp to the appropriate section of the
file.
- Create `gen<OpName>Clauses()` functions containing the clause processing code
specific for the associated OpenMP construct.
- Update `gen<OpName>Op()` functions to call the corresponding
`gen<OpName>Clauses()` function.
- Sort calls to `ClauseProcessor::process<ClauseName>()` alphabetically, to
avoid inadvertently relying on some arbitrary order. Update some tests that
broke due to the order change.
- Normalize `genOMP()` functions so they all delegate the generation of MLIR to
`gen<OpName>Op()` functions following the same pattern.
- Only process `nowait` clause on `TARGET` constructs if not compiling for the
target device.
A later patch can move the calls to `gen<OpName>Clauses()` out of
`gen<OpName>Op()` functions and passing completed clause structures instead, in
preparation to supporting composite constructs. That will make it possible to
reuse clause processing for a given leaf construct when appearing alone or in a
combined or composite construct, while controlling where the associated code is
produced.
---
flang/lib/Lower/OpenMP/ClauseProcessor.cpp | 4 +-
flang/lib/Lower/OpenMP/ClauseProcessor.h | 3 +-
flang/lib/Lower/OpenMP/OpenMP.cpp | 2090 +++++++++--------
flang/test/Lower/OpenMP/FIR/target.f90 | 2 +-
flang/test/Lower/OpenMP/target.f90 | 2 +-
.../use-device-ptr-to-use-device-addr.f90 | 4 +-
6 files changed, 1173 insertions(+), 932 deletions(-)
diff --git a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
index ee1f6c2fbc7e89..e2b26b3025049f 100644
--- a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
+++ b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
@@ -804,8 +804,8 @@ createMapInfoOp(fir::FirOpBuilder &builder, mlir::Location loc,
}
bool ClauseProcessor::processMap(
- mlir::Location currentLocation, const llvm::omp::Directive &directive,
- Fortran::lower::StatementContext &stmtCtx, mlir::omp::MapClauseOps &result,
+ mlir::Location currentLocation, Fortran::lower::StatementContext &stmtCtx,
+ mlir::omp::MapClauseOps &result,
llvm::SmallVectorImpl<const Fortran::semantics::Symbol *> *mapSyms,
llvm::SmallVectorImpl<mlir::Location> *mapSymLocs,
llvm::SmallVectorImpl<mlir::Type> *mapSymTypes) const {
diff --git a/flang/lib/Lower/OpenMP/ClauseProcessor.h b/flang/lib/Lower/OpenMP/ClauseProcessor.h
index d933e0a913d2bc..9e59d754280ef4 100644
--- a/flang/lib/Lower/OpenMP/ClauseProcessor.h
+++ b/flang/lib/Lower/OpenMP/ClauseProcessor.h
@@ -102,8 +102,7 @@ class ClauseProcessor {
// They may be used later on to create the block_arguments for some of the
// target directives that require it.
bool processMap(
- mlir::Location currentLocation, const llvm::omp::Directive &directive,
- Fortran::lower::StatementContext &stmtCtx,
+ mlir::Location currentLocation, Fortran::lower::StatementContext &stmtCtx,
mlir::omp::MapClauseOps &result,
llvm::SmallVectorImpl<const Fortran::semantics::Symbol *> *mapSyms =
nullptr,
diff --git a/flang/lib/Lower/OpenMP/OpenMP.cpp b/flang/lib/Lower/OpenMP/OpenMP.cpp
index d67060d1cce72b..b6de2079a973f5 100644
--- a/flang/lib/Lower/OpenMP/OpenMP.cpp
+++ b/flang/lib/Lower/OpenMP/OpenMP.cpp
@@ -237,6 +237,276 @@ createAndSetPrivatizedLoopVar(Fortran::lower::AbstractConverter &converter,
return storeOp;
}
+// This helper function implements the functionality of "promoting"
+// non-CPTR arguments of use_device_ptr to use_device_addr
+// arguments (automagic conversion of use_device_ptr ->
+// use_device_addr in these cases). The way we do so currently is
+// through the shuffling of operands from the devicePtrOperands to
+// deviceAddrOperands where neccesary and re-organizing the types,
+// locations and symbols to maintain the correct ordering of ptr/addr
+// input -> BlockArg.
+//
+// This effectively implements some deprecated OpenMP functionality
+// that some legacy applications unfortunately depend on
+// (deprecated in specification version 5.2):
+//
+// "If a list item in a use_device_ptr clause is not of type C_PTR,
+// the behavior is as if the list item appeared in a use_device_addr
+// clause. Support for such list items in a use_device_ptr clause
+// is deprecated."
+static void promoteNonCPtrUseDevicePtrArgsToUseDeviceAddr(
+ mlir::omp::UseDeviceClauseOps &clauseOps,
+ llvm::SmallVectorImpl<mlir::Type> &useDeviceTypes,
+ llvm::SmallVectorImpl<mlir::Location> &useDeviceLocs,
+ llvm::SmallVectorImpl<const Fortran::semantics::Symbol *>
+ &useDeviceSymbols) {
+ auto moveElementToBack = [](size_t idx, auto &vector) {
+ auto *iter = std::next(vector.begin(), idx);
+ vector.push_back(*iter);
+ vector.erase(iter);
+ };
+
+ // Iterate over our use_device_ptr list and shift all non-cptr arguments into
+ // use_device_addr.
+ for (auto *it = clauseOps.useDevicePtrVars.begin();
+ it != clauseOps.useDevicePtrVars.end();) {
+ if (!fir::isa_builtin_cptr_type(fir::unwrapRefType(it->getType()))) {
+ clauseOps.useDeviceAddrVars.push_back(*it);
+ // We have to shuffle the symbols around as well, to maintain
+ // the correct Input -> BlockArg for use_device_ptr/use_device_addr.
+ // NOTE: However, as map's do not seem to be included currently
+ // this isn't as pertinent, but we must try to maintain for
+ // future alterations. I believe the reason they are not currently
+ // is that the BlockArg assign/lowering needs to be extended
+ // to a greater set of types.
+ auto idx = std::distance(clauseOps.useDevicePtrVars.begin(), it);
+ moveElementToBack(idx, useDeviceTypes);
+ moveElementToBack(idx, useDeviceLocs);
+ moveElementToBack(idx, useDeviceSymbols);
+ it = clauseOps.useDevicePtrVars.erase(it);
+ continue;
+ }
+ ++it;
+ }
+}
+
+/// Extract the list of function and variable symbols affected by the given
+/// 'declare target' directive and return the intended device type for them.
+static void getDeclareTargetInfo(
+ Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval,
+ const Fortran::parser::OpenMPDeclareTargetConstruct &declareTargetConstruct,
+ mlir::omp::DeclareTargetClauseOps &clauseOps,
+ llvm::SmallVectorImpl<DeclareTargetCapturePair> &symbolAndClause) {
+ const auto &spec = std::get<Fortran::parser::OmpDeclareTargetSpecifier>(
+ declareTargetConstruct.t);
+ if (const auto *objectList{
+ Fortran::parser::Unwrap<Fortran::parser::OmpObjectList>(spec.u)}) {
+ ObjectList objects{makeList(*objectList, semaCtx)};
+ // Case: declare target(func, var1, var2)
+ gatherFuncAndVarSyms(objects, mlir::omp::DeclareTargetCaptureClause::to,
+ symbolAndClause);
+ } else if (const auto *clauseList{
+ Fortran::parser::Unwrap<Fortran::parser::OmpClauseList>(
+ spec.u)}) {
+ if (clauseList->v.empty()) {
+ // Case: declare target, implicit capture of function
+ symbolAndClause.emplace_back(
+ mlir::omp::DeclareTargetCaptureClause::to,
+ eval.getOwningProcedure()->getSubprogramSymbol());
+ }
+
+ ClauseProcessor cp(converter, semaCtx, *clauseList);
+ cp.processDeviceType(clauseOps);
+ cp.processEnter(symbolAndClause);
+ cp.processLink(symbolAndClause);
+ cp.processTo(symbolAndClause);
+ cp.processTODO<clause::Indirect>(converter.getCurrentLocation(),
+ llvm::omp::Directive::OMPD_declare_target);
+ }
+}
+
+static void collectDeferredDeclareTargets(
+ Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval,
+ const Fortran::parser::OpenMPDeclareTargetConstruct &declareTargetConstruct,
+ llvm::SmallVectorImpl<Fortran::lower::OMPDeferredDeclareTargetInfo>
+ &deferredDeclareTarget) {
+ mlir::omp::DeclareTargetClauseOps clauseOps;
+ llvm::SmallVector<DeclareTargetCapturePair> symbolAndClause;
+ getDeclareTargetInfo(converter, semaCtx, eval, declareTargetConstruct,
+ clauseOps, symbolAndClause);
+ // Return the device type only if at least one of the targets for the
+ // directive is a function or subroutine
+ mlir::ModuleOp mod = converter.getFirOpBuilder().getModule();
+
+ for (const DeclareTargetCapturePair &symClause : symbolAndClause) {
+ mlir::Operation *op = mod.lookupSymbol(converter.mangleName(
+ std::get<const Fortran::semantics::Symbol &>(symClause)));
+
+ if (!op) {
+ deferredDeclareTarget.push_back({std::get<0>(symClause),
+ clauseOps.deviceType,
+ std::get<1>(symClause)});
+ }
+ }
+}
+
+static std::optional<mlir::omp::DeclareTargetDeviceType>
+getDeclareTargetFunctionDevice(
+ Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval,
+ const Fortran::parser::OpenMPDeclareTargetConstruct
+ &declareTargetConstruct) {
+ mlir::omp::DeclareTargetClauseOps clauseOps;
+ llvm::SmallVector<DeclareTargetCapturePair> symbolAndClause;
+ getDeclareTargetInfo(converter, semaCtx, eval, declareTargetConstruct,
+ clauseOps, symbolAndClause);
+
+ // Return the device type only if at least one of the targets for the
+ // directive is a function or subroutine
+ mlir::ModuleOp mod = converter.getFirOpBuilder().getModule();
+ for (const DeclareTargetCapturePair &symClause : symbolAndClause) {
+ mlir::Operation *op = mod.lookupSymbol(converter.mangleName(
+ std::get<const Fortran::semantics::Symbol &>(symClause)));
+
+ if (mlir::isa_and_nonnull<mlir::func::FuncOp>(op))
+ return clauseOps.deviceType;
+ }
+
+ return std::nullopt;
+}
+
+static llvm::SmallVector<const Fortran::semantics::Symbol *>
+genLoopVars(mlir::Operation *op, Fortran::lower::AbstractConverter &converter,
+ mlir::Location &loc,
+ llvm::ArrayRef<const Fortran::semantics::Symbol *> args) {
+ fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
+ auto ®ion = op->getRegion(0);
+
+ std::size_t loopVarTypeSize = 0;
+ for (const Fortran::semantics::Symbol *arg : args)
+ loopVarTypeSize = std::max(loopVarTypeSize, arg->GetUltimate().size());
+ mlir::Type loopVarType = getLoopVarType(converter, loopVarTypeSize);
+ llvm::SmallVector<mlir::Type> tiv(args.size(), loopVarType);
+ llvm::SmallVector<mlir::Location> locs(args.size(), loc);
+ firOpBuilder.createBlock(®ion, {}, tiv, locs);
+ // The argument is not currently in memory, so make a temporary for the
+ // argument, and store it there, then bind that location to the argument.
+ mlir::Operation *storeOp = nullptr;
+ for (auto [argIndex, argSymbol] : llvm::enumerate(args)) {
+ mlir::Value indexVal = fir::getBase(region.front().getArgument(argIndex));
+ storeOp =
+ createAndSetPrivatizedLoopVar(converter, loc, indexVal, argSymbol);
+ }
+ firOpBuilder.setInsertionPointAfter(storeOp);
+
+ return llvm::SmallVector<const Fortran::semantics::Symbol *>(args);
+}
+
+static void genReductionVars(
+ mlir::Operation *op, Fortran::lower::AbstractConverter &converter,
+ mlir::Location &loc,
+ llvm::ArrayRef<const Fortran::semantics::Symbol *> reductionArgs,
+ llvm::ArrayRef<mlir::Type> reductionTypes) {
+ fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
+ llvm::SmallVector<mlir::Location> blockArgLocs(reductionArgs.size(), loc);
+
+ mlir::Block *entryBlock = firOpBuilder.createBlock(
+ &op->getRegion(0), {}, reductionTypes, blockArgLocs);
+
+ // Bind the reduction arguments to their block arguments.
+ for (auto [arg, prv] :
+ llvm::zip_equal(reductionArgs, entryBlock->getArguments())) {
+ converter.bindSymbol(*arg, prv);
+ }
+}
+
+static llvm::SmallVector<const Fortran::semantics::Symbol *>
+genLoopAndReductionVars(
+ mlir::Operation *op, Fortran::lower::AbstractConverter &converter,
+ mlir::Location &loc,
+ llvm::ArrayRef<const Fortran::semantics::Symbol *> loopArgs,
+ llvm::ArrayRef<const Fortran::semantics::Symbol *> reductionArgs,
+ llvm::ArrayRef<mlir::Type> reductionTypes) {
+ fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
+
+ llvm::SmallVector<mlir::Type> blockArgTypes;
+ llvm::SmallVector<mlir::Location> blockArgLocs;
+ blockArgTypes.reserve(loopArgs.size() + reductionArgs.size());
+ blockArgLocs.reserve(blockArgTypes.size());
+ mlir::Block *entryBlock;
+
+ if (loopArgs.size()) {
+ std::size_t loopVarTypeSize = 0;
+ for (const Fortran::semantics::Symbol *arg : loopArgs)
+ loopVarTypeSize = std::max(loopVarTypeSize, arg->GetUltimate().size());
+ mlir::Type loopVarType = getLoopVarType(converter, loopVarTypeSize);
+ std::fill_n(std::back_inserter(blockArgTypes), loopArgs.size(),
+ loopVarType);
+ std::fill_n(std::back_inserter(blockArgLocs), loopArgs.size(), loc);
+ }
+ if (reductionArgs.size()) {
+ llvm::copy(reductionTypes, std::back_inserter(blockArgTypes));
+ std::fill_n(std::back_inserter(blockArgLocs), reductionArgs.size(), loc);
+ }
+ entryBlock = firOpBuilder.createBlock(&op->getRegion(0), {}, blockArgTypes,
+ blockArgLocs);
+ // The argument is not currently in memory, so make a temporary for the
+ // argument, and store it there, then bind that location to the argument.
+ if (loopArgs.size()) {
+ mlir::Operation *storeOp = nullptr;
+ for (auto [argIndex, argSymbol] : llvm::enumerate(loopArgs)) {
+ mlir::Value indexVal =
+ fir::getBase(op->getRegion(0).front().getArgument(argIndex));
+ storeOp =
+ createAndSetPrivatizedLoopVar(converter, loc, indexVal, argSymbol);
+ }
+ firOpBuilder.setInsertionPointAfter(storeOp);
+ }
+ // Bind the reduction arguments to their block arguments
+ for (auto [arg, prv] : llvm::zip_equal(
+ reductionArgs,
+ llvm::drop_begin(entryBlock->getArguments(), loopArgs.size()))) {
+ converter.bindSymbol(*arg, prv);
+ }
+
+ return llvm::SmallVector<const Fortran::semantics::Symbol *>(loopArgs);
+}
+
+static void
+markDeclareTarget(mlir::Operation *op,
+ Fortran::lower::AbstractConverter &converter,
+ mlir::omp::DeclareTargetCaptureClause captureClause,
+ mlir::omp::DeclareTargetDeviceType deviceType) {
+ // TODO: Add support for program local variables with declare target applied
+ auto declareTargetOp = llvm::dyn_cast<mlir::omp::DeclareTargetInterface>(op);
+ if (!declareTargetOp)
+ fir::emitFatalError(
+ converter.getCurrentLocation(),
+ "Attempt to apply declare target on unsupported operation");
+
+ // The function or global already has a declare target applied to it, very
+ // likely through implicit capture (usage in another declare target
+ // function/subroutine). It should be marked as any if it has been assigned
+ // both host and nohost, else we skip, as there is no change
+ if (declareTargetOp.isDeclareTarget()) {
+ if (declareTargetOp.getDeclareTargetDeviceType() != deviceType)
+ declareTargetOp.setDeclareTarget(mlir::omp::DeclareTargetDeviceType::any,
+ captureClause);
+ return;
+ }
+
+ declareTargetOp.setDeclareTarget(deviceType, captureClause);
+}
+
+//===----------------------------------------------------------------------===//
+// Op body generation helper structures and functions
+//===----------------------------------------------------------------------===//
+
struct OpWithBodyGenInfo {
/// A type for a code-gen callback function. This takes as argument the op for
/// which the code is being generated and returns the arguments of the op's
@@ -508,543 +778,726 @@ static void genBodyOfTargetDataOp(
genNestedEvaluations(converter, eval);
}
-template <typename OpTy, typename... Args>
-static OpTy genOpWithBody(OpWithBodyGenInfo &info, Args &&...args) {
- auto op = info.converter.getFirOpBuilder().create<OpTy>(
- info.loc, std::forward<Args>(args)...);
- createBodyOfOp<OpTy>(op, info);
- return op;
-}
-
-static mlir::omp::MasterOp
-genMasterOp(Fortran::lower::AbstractConverter &converter,
- Fortran::semantics::SemanticsContext &semaCtx,
- Fortran::lower::pft::Evaluation &eval, bool genNested,
- mlir::Location currentLocation) {
- return genOpWithBody<mlir::omp::MasterOp>(
- OpWithBodyGenInfo(converter, semaCtx, currentLocation, eval)
- .setGenNested(genNested));
-}
-
-static mlir::omp::OrderedRegionOp
-genOrderedRegionOp(Fortran::lower::AbstractConverter &converter,
- Fortran::semantics::SemanticsContext &semaCtx,
- Fortran::lower::pft::Evaluation &eval, bool genNested,
- mlir::Location currentLocation,
- const Fortran::parser::OmpClauseList &clauseList) {
- mlir::omp::OrderedRegionClauseOps clauseOps;
+// This functions creates a block for the body of the targetOp's region. It adds
+// all the symbols present in mapSymbols as block arguments to this block.
+static void
+genBodyOfTargetOp(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval, bool genNested,
+ mlir::omp::TargetOp &targetOp,
+ llvm::ArrayRef<const Fortran::semantics::Symbol *> mapSyms,
+ llvm::ArrayRef<mlir::Location> mapSymLocs,
+ llvm::ArrayRef<mlir::Type> mapSymTypes,
+ const mlir::Location ¤tLocation) {
+ assert(mapSymTypes.size() == mapSymLocs.size());
- ClauseProcessor cp(converter, semaCtx, clauseList);
- cp.processTODO<clause::Simd>(currentLocation,
- llvm::omp::Directive::OMPD_ordered);
+ fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
+ mlir::Region ®ion = targetOp.getRegion();
- return genOpWithBody<mlir::omp::OrderedRegionOp>(
- OpWithBodyGenInfo(converter, semaCtx, currentLocation, eval)
- .setGenNested(genNested),
- clauseOps);
-}
+ auto *regionBlock =
+ firOpBuilder.createBlock(®ion, {}, mapSymTypes, mapSymLocs);
-static mlir::omp::ParallelOp
-genParallelOp(Fortran::lower::AbstractConverter &converter,
- Fortran::lower::SymMap &symTable,
- Fortran::semantics::SemanticsContext &semaCtx,
- Fortran::lower::pft::Evaluation &eval, bool genNested,
- mlir::Location currentLocation,
- const Fortran::parser::OmpClauseList &clauseList,
- bool outerCombined = false) {
- fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
- Fortran::lower::StatementContext stmtCtx;
- mlir::omp::ParallelClauseOps clauseOps;
- llvm::SmallVector<const Fortran::semantics::Symbol *> privateSyms;
- llvm::SmallVector<mlir::Type> reductionTypes;
- llvm::SmallVector<const Fortran::semantics::Symbol *> reductionSyms;
+ // Clones the `bounds` placing them inside the target region and returns them.
+ auto cloneBound = [&](mlir::Value bound) {
+ if (mlir::isMemoryEffectFree(bound.getDefiningOp())) {
+ mlir::Operation *clonedOp = bound.getDefiningOp()->clone();
+ regionBlock->push_back(clonedOp);
+ return clonedOp->getResult(0);
+ }
+ TODO(converter.getCurrentLocation(),
+ "target map clause operand unsupported bound type");
+ };
- ClauseProcessor cp(converter, semaCtx, clauseList);
- cp.processIf(llvm::omp::Directive::OMPD_parallel, clauseOps);
- cp.processNumThreads(stmtCtx, clauseOps);
- cp.processProcBind(clauseOps);
- cp.processDefault();
- cp.processAllocate(clauseOps);
+ auto cloneBounds = [cloneBound](llvm::ArrayRef<mlir::Value> bounds) {
+ llvm::SmallVector<mlir::Value> clonedBounds;
+ for (mlir::Value bound : bounds)
+ clonedBounds.emplace_back(cloneBound(bound));
+ return clonedBounds;
+ };
- if (!outerCombined)
- cp.processReduction(currentLocation, clauseOps, &reductionTypes,
- &reductionSyms);
+ // Bind the symbols to their corresponding block arguments.
+ for (auto [argIndex, argSymbol] : llvm::enumerate(mapSyms)) {
+ const mlir::BlockArgument &arg = region.getArgument(argIndex);
+ // Avoid capture of a reference to a structured binding.
+ const Fortran::semantics::Symbol *sym = argSymbol;
+ // Structure component symbols don't have bindings.
+ if (sym->owner().IsDerivedType())
+ continue;
+ fir::ExtendedValue extVal = converter.getSymbolExtendedValue(*sym);
+ extVal.match(
+ [&](const fir::BoxValue &v) {
+ converter.bindSymbol(*sym,
+ fir::BoxValue(arg, cloneBounds(v.getLBounds()),
+ v.getExplicitParameters(),
+ v.getExplicitExtents()));
+ },
+ [&](const fir::MutableBoxValue &v) {
+ converter.bindSymbol(
+ *sym, fir::MutableBoxValue(arg, cloneBounds(v.getLBounds()),
+ v.getMutableProperties()));
+ },
+ [&](const fir::ArrayBoxValue &v) {
+ converter.bindSymbol(
+ *sym, fir::ArrayBoxValue(arg, cloneBounds(v.getExtents()),
+ cloneBounds(v.getLBounds()),
+ v.getSourceBox()));
+ },
+ [&](const fir::CharArrayBoxValue &v) {
+ converter.bindSymbol(
+ *sym, fir::CharArrayBoxValue(arg, cloneBound(v.getLen()),
+ cloneBounds(v.getExtents()),
+ cloneBounds(v.getLBounds())));
+ },
+ [&](const fir::CharBoxValue &v) {
+ converter.bindSymbol(*sym,
+ fir::CharBoxValue(arg, cloneBound(v.getLen())));
+ },
+ [&](const fir::UnboxedValue &v) { converter.bindSymbol(*sym, arg); },
+ [&](const auto &) {
+ TODO(converter.getCurrentLocation(),
+ "target map clause operand unsupported type");
+ });
+ }
- if (ReductionProcessor::doReductionByRef(clauseOps.reductionVars))
- clauseOps.reductionByRefAttr = firOpBuilder.getUnitAttr();
+ // Check if cloning the bounds introduced any dependency on the outer region.
+ // If so, then either clone them as well if they are MemoryEffectFree, or else
+ // copy them to a new temporary and add them to the map and block_argument
+ // lists and replace their uses with the new temporary.
+ llvm::SetVector<mlir::Value> valuesDefinedAbove;
+ mlir::getUsedValuesDefinedAbove(region, valuesDefinedAbove);
+ while (!valuesDefinedAbove.empty()) {
+ for (mlir::Value val : valuesDefinedAbove) {
+ mlir::Operation *valOp = val.getDefiningOp();
+ if (mlir::isMemoryEffectFree(valOp)) {
+ mlir::Operation *clonedOp = valOp->clone();
+ regionBlock->push_front(clonedOp);
+ val.replaceUsesWithIf(
+ clonedOp->getResult(0), [regionBlock](mlir::OpOperand &use) {
+ return use.getOwner()->getBlock() == regionBlock;
+ });
+ } else {
+ auto savedIP = firOpBuilder.getInsertionPoint();
+ firOpBuilder.setInsertionPointAfter(valOp);
+ auto copyVal =
+ firOpBuilder.createTemporary(val.getLoc(), val.getType());
+ firOpBuilder.createStoreWithConvert(copyVal.getLoc(), val, copyVal);
- auto reductionCallback = [&](mlir::Operation *op) {
- llvm::SmallVector<mlir::Location> locs(clauseOps.reductionVars.size(),
- currentLocation);
- auto *block =
- firOpBuilder.createBlock(&op->getRegion(0), {}, reductionTypes, locs);
- for (auto [arg, prv] :
- llvm::zip_equal(reductionSyms, block->getArguments())) {
- converter.bindSymbol(*arg, prv);
+ llvm::SmallVector<mlir::Value> bounds;
+ std::stringstream name;
+ firOpBuilder.setInsertionPoint(targetOp);
+ mlir::Value mapOp = createMapInfoOp(
+ firOpBuilder, copyVal.getLoc(), copyVal, mlir::Value{}, name.str(),
+ bounds, llvm::SmallVector<mlir::Value>{},
+ static_cast<
+ std::underlying_type_t<llvm::omp::OpenMPOffloadMappingFlags>>(
+ llvm::omp::OpenMPOffloadMappingFlags::OMP_MAP_IMPLICIT),
+ mlir::omp::VariableCaptureKind::ByCopy, copyVal.getType());
+ targetOp.getMapOperandsMutable().append(mapOp);
+ mlir::Value clonedValArg =
+ region.addArgument(copyVal.getType(), copyVal.getLoc());
+ firOpBuilder.setInsertionPointToStart(regionBlock);
+ auto loadOp = firOpBuilder.create<fir::LoadOp>(clonedValArg.getLoc(),
+ clonedValArg);
+ val.replaceUsesWithIf(
+ loadOp->getResult(0), [regionBlock](mlir::OpOperand &use) {
+ return use.getOwner()->getBlock() == regionBlock;
+ });
+ firOpBuilder.setInsertionPoint(regionBlock, savedIP);
+ }
}
- return reductionSyms;
- };
-
- OpWithBodyGenInfo genInfo =
- OpWithBodyGenInfo(converter, semaCtx, currentLocation, eval)
- .setGenNested(genNested)
- .setOuterCombined(outerCombined)
- .setClauses(&clauseList)
- .setReductions(&reductionSyms, &reductionTypes)
- .setGenRegionEntryCb(reductionCallback);
+ valuesDefinedAbove.clear();
+ mlir::getUsedValuesDefinedAbove(region, valuesDefinedAbove);
+ }
- if (!enableDelayedPrivatization)
- return genOpWithBody<mlir::omp::ParallelOp>(genInfo, clauseOps);
+ // Insert dummy instruction to remember the insertion position. The
+ // marker will be deleted since there are not uses.
+ // In the HLFIR flow there are hlfir.declares inserted above while
+ // setting block arguments.
+ mlir::Value undefMarker = firOpBuilder.create<fir::UndefOp>(
+ targetOp.getOperation()->getLoc(), firOpBuilder.getIndexType());
- bool privatize = !outerCombined;
- DataSharingProcessor dsp(converter, semaCtx, clauseList, eval,
- /*useDelayedPrivatization=*/true, &symTable);
+ // Create blocks for unstructured regions. This has to be done since
+ // blocks are initially allocated with the function as the parent region.
+ if (eval.lowerAsUnstructured()) {
+ Fortran::lower::createEmptyRegionBlocks<mlir::omp::TerminatorOp,
+ mlir::omp::YieldOp>(
+ firOpBuilder, eval.getNestedEvaluations());
+ }
- if (privatize)
- dsp.processStep1(&clauseOps, &privateSyms);
+ firOpBuilder.create<mlir::omp::TerminatorOp>(currentLocation);
- auto genRegionEntryCB = [&](mlir::Operation *op) {
- auto parallelOp = llvm::cast<mlir::omp::ParallelOp>(op);
+ // Create the insertion point after the marker.
+ firOpBuilder.setInsertionPointAfter(undefMarker.getDefiningOp());
+ if (genNested)
+ genNestedEvaluations(converter, eval);
+}
- llvm::SmallVector<mlir::Location> reductionLocs(
- clauseOps.reductionVars.size(), currentLocation);
+template <typename OpTy, typename... Args>
+static OpTy genOpWithBody(OpWithBodyGenInfo &info, Args &&...args) {
+ auto op = info.converter.getFirOpBuilder().create<OpTy>(
+ info.loc, std::forward<Args>(args)...);
+ createBodyOfOp<OpTy>(op, info);
+ return op;
+}
- mlir::OperandRange privateVars = parallelOp.getPrivateVars();
- mlir::Region ®ion = parallelOp.getRegion();
+//===----------------------------------------------------------------------===//
+// Code generation functions for clauses
+//===----------------------------------------------------------------------===//
- llvm::SmallVector<mlir::Type> privateVarTypes = reductionTypes;
- privateVarTypes.reserve(privateVarTypes.size() + privateVars.size());
- llvm::transform(privateVars, std::back_inserter(privateVarTypes),
- [](mlir::Value v) { return v.getType(); });
+static void genCriticalDeclareClauses(
+ Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ const Fortran::parser::OmpClauseList &clauses, mlir::Location loc,
+ mlir::omp::CriticalClauseOps &clauseOps, llvm::StringRef name) {
+ ClauseProcessor cp(converter, semaCtx, clauses);
+ cp.processHint(clauseOps);
+ clauseOps.nameAttr =
+ mlir::StringAttr::get(converter.getFirOpBuilder().getContext(), name);
+}
- llvm::SmallVector<mlir::Location> privateVarLocs = reductionLocs;
- privateVarLocs.reserve(privateVarLocs.size() + privateVars.size());
- llvm::transform(privateVars, std::back_inserter(privateVarLocs),
- [](mlir::Value v) { return v.getLoc(); });
+static void genFlushClauses(
+ Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ const std::optional<Fortran::parser::OmpObjectList> &objects,
+ const std::optional<std::list<Fortran::parser::OmpMemoryOrderClause>>
+ &clauses,
+ mlir::Location loc, llvm::SmallVectorImpl<mlir::Value> &operandRange) {
+ if (objects)
+ genObjectList2(*objects, converter, operandRange);
+
+ if (clauses && clauses->size() > 0)
+ TODO(converter.getCurrentLocation(), "Handle OmpMemoryOrderClause");
+}
- firOpBuilder.createBlock(®ion, /*insertPt=*/{}, privateVarTypes,
- privateVarLocs);
+static void
+genOrderedRegionClauses(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ const Fortran::parser::OmpClauseList &clauses,
+ mlir::Location loc,
+ mlir::omp::OrderedRegionClauseOps &clauseOps) {
+ ClauseProcessor cp(converter, semaCtx, clauses);
+ cp.processTODO<clause::Simd>(loc, llvm::omp::Directive::OMPD_ordered);
+}
- llvm::SmallVector<const Fortran::semantics::Symbol *> allSymbols =
- reductionSyms;
- allSymbols.append(privateSyms);
- for (auto [arg, prv] : llvm::zip_equal(allSymbols, region.getArguments())) {
- converter.bindSymbol(*arg, prv);
- }
+static void genParallelClauses(
+ Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::StatementContext &stmtCtx,
+ const Fortran::parser::OmpClauseList &clauses, mlir::Location loc,
+ bool processReduction, mlir::omp::ParallelClauseOps &clauseOps,
+ llvm::SmallVectorImpl<mlir::Type> &reductionTypes,
+ llvm::SmallVectorImpl<const Fortran::semantics::Symbol *> &reductionSyms) {
+ ClauseProcessor cp(converter, semaCtx, clauses);
+ cp.processAllocate(clauseOps);
+ cp.processDefault();
+ cp.processIf(llvm::omp::Directive::OMPD_parallel, clauseOps);
+ cp.processProcBind(clauseOps);
- return allSymbols;
- };
+ if (processReduction) {
+ cp.processReduction(loc, clauseOps, &reductionTypes, &reductionSyms);
+ if (ReductionProcessor::doReductionByRef(clauseOps.reductionVars))
+ clauseOps.reductionByRefAttr = converter.getFirOpBuilder().getUnitAttr();
+ }
- // TODO Merge with the reduction CB.
- genInfo.setGenRegionEntryCb(genRegionEntryCB).setDataSharingProcessor(&dsp);
- return genOpWithBody<mlir::omp::ParallelOp>(genInfo, clauseOps);
+ cp.processNumThreads(stmtCtx, clauseOps);
}
-static mlir::omp::SectionOp
-genSectionOp(Fortran::lower::AbstractConverter &converter,
- Fortran::semantics::SemanticsContext &semaCtx,
- Fortran::lower::pft::Evaluation &eval, bool genNested,
- mlir::Location currentLocation,
- const Fortran::parser::OmpClauseList §ionsClauseList) {
- // Currently only private/firstprivate clause is handled, and
- // all privatization is done within `omp.section` operations.
- return genOpWithBody<mlir::omp::SectionOp>(
- OpWithBodyGenInfo(converter, semaCtx, currentLocation, eval)
- .setGenNested(genNested)
- .setClauses(§ionsClauseList));
+static void genSectionsClauses(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ const Fortran::parser::OmpClauseList &clauses,
+ mlir::Location loc,
+ bool clausesFromBeginSections,
+ mlir::omp::SectionsClauseOps &clauseOps) {
+ ClauseProcessor cp(converter, semaCtx, clauses);
+ if (clausesFromBeginSections) {
+ cp.processAllocate(clauseOps);
+ cp.processSectionsReduction(loc, clauseOps);
+ // TODO Support delayed privatization.
+ } else {
+ cp.processNowait(clauseOps);
+ }
}
-static mlir::omp::SingleOp
-genSingleOp(Fortran::lower::AbstractConverter &converter,
- Fortran::semantics::SemanticsContext &semaCtx,
- Fortran::lower::pft::Evaluation &eval, bool genNested,
- mlir::Location currentLocation,
- const Fortran::parser::OmpClauseList &beginClauseList,
- const Fortran::parser::OmpClauseList &endClauseList) {
- mlir::omp::SingleClauseOps clauseOps;
+static void genSimdLoopClauses(
+ Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::StatementContext &stmtCtx,
+ Fortran::lower::pft::Evaluation &eval,
+ const Fortran::parser::OmpClauseList &clauses, mlir::Location loc,
+ mlir::omp::SimdLoopClauseOps &clauseOps,
+ llvm::SmallVectorImpl<const Fortran::semantics::Symbol *> &iv) {
+ ClauseProcessor cp(converter, semaCtx, clauses);
+ cp.processCollapse(loc, eval, clauseOps, iv);
+ cp.processIf(llvm::omp::Directive::OMPD_simd, clauseOps);
+ cp.processReduction(loc, clauseOps);
+ cp.processSafelen(clauseOps);
+ cp.processSimdlen(clauseOps);
+ clauseOps.loopInclusiveAttr = converter.getFirOpBuilder().getUnitAttr();
+ // TODO Support delayed privatization.
- ClauseProcessor cp(converter, semaCtx, beginClauseList);
- cp.processAllocate(clauseOps);
+ cp.processTODO<clause::Aligned, clause::Allocate, clause::Linear,
+ clause::Nontemporal, clause::Order>(
+ loc, llvm::omp::Directive::OMPD_simd);
+}
+
+static void genSingleClauses(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ const Fortran::parser::OmpClauseList &beginClauses,
+ const Fortran::parser::OmpClauseList &endClauses,
+ mlir::Location loc,
+ mlir::omp::SingleClauseOps &clauseOps) {
+ ClauseProcessor bcp(converter, semaCtx, beginClauses);
+ bcp.processAllocate(clauseOps);
// TODO Support delayed privatization.
- ClauseProcessor ecp(converter, semaCtx, endClauseList);
+ ClauseProcessor ecp(converter, semaCtx, endClauses);
+ ecp.processCopyprivate(loc, clauseOps);
ecp.processNowait(clauseOps);
- ecp.processCopyprivate(currentLocation, clauseOps);
+}
- return genOpWithBody<mlir::omp::SingleOp>(
- OpWithBodyGenInfo(converter, semaCtx, currentLocation, eval)
- .setGenNested(genNested)
- .setClauses(&beginClauseList),
- clauseOps);
+static void genTargetClauses(
+ Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::StatementContext &stmtCtx,
+ const Fortran::parser::OmpClauseList &clauses, mlir::Location loc,
+ bool processHostOnlyClauses, bool processReduction,
+ mlir::omp::TargetClauseOps &clauseOps,
+ llvm::SmallVectorImpl<const Fortran::semantics::Symbol *> &mapSyms,
+ llvm::SmallVectorImpl<mlir::Location> &mapSymLocs,
+ llvm::SmallVectorImpl<mlir::Type> &mapSymTypes) {
+ ClauseProcessor cp(converter, semaCtx, clauses);
+ cp.processDepend(clauseOps);
+ cp.processDevice(stmtCtx, clauseOps);
+ cp.processIf(llvm::omp::Directive::OMPD_target, clauseOps);
+ cp.processMap(loc, stmtCtx, clauseOps, &mapSyms, &mapSymLocs, &mapSymTypes);
+ cp.processThreadLimit(stmtCtx, clauseOps);
+ // TODO Support delayed privatization.
+
+ if (processHostOnlyClauses)
+ cp.processNowait(clauseOps);
+
+ cp.processTODO<clause::Allocate, clause::Defaultmap, clause::Firstprivate,
+ clause::HasDeviceAddr, clause::InReduction,
+ clause::IsDevicePtr, clause::Private, clause::Reduction,
+ clause::UsesAllocators>(loc,
+ llvm::omp::Directive::OMPD_target);
}
-static mlir::omp::TaskOp
-genTaskOp(Fortran::lower::AbstractConverter &converter,
- Fortran::semantics::SemanticsContext &semaCtx,
- Fortran::lower::pft::Evaluation &eval, bool genNested,
- mlir::Location currentLocation,
- const Fortran::parser::OmpClauseList &clauseList) {
- Fortran::lower::StatementContext stmtCtx;
- mlir::omp::TaskClauseOps clauseOps;
+static void genTargetDataClauses(
+ Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::StatementContext &stmtCtx,
+ const Fortran::parser::OmpClauseList &clauses, mlir::Location loc,
+ mlir::omp::TargetDataClauseOps &clauseOps,
+ llvm::SmallVectorImpl<mlir::Type> &useDeviceTypes,
+ llvm::SmallVectorImpl<mlir::Location> &useDeviceLocs,
+ llvm::SmallVectorImpl<const Fortran::semantics::Symbol *> &useDeviceSyms) {
+ ClauseProcessor cp(converter, semaCtx, clauses);
+ cp.processDevice(stmtCtx, clauseOps);
+ cp.processIf(llvm::omp::Directive::OMPD_target_data, clauseOps);
+ cp.processMap(loc, stmtCtx, clauseOps);
+ cp.processUseDeviceAddr(clauseOps, useDeviceTypes, useDeviceLocs,
+ useDeviceSyms);
+ cp.processUseDevicePtr(clauseOps, useDeviceTypes, useDeviceLocs,
+ useDeviceSyms);
- ClauseProcessor cp(converter, semaCtx, clauseList);
- cp.processIf(llvm::omp::Directive::OMPD_task, clauseOps);
+ // This function implements the deprecated functionality of use_device_ptr
+ // that allows users to provide non-CPTR arguments to it with the caveat
+ // that the compiler will treat them as use_device_addr. A lot of legacy
+ // code may still depend on this functionality, so we should support it
+ // in some manner. We do so currently by simply shifting non-cptr operands
+ // from the use_device_ptr list into the front of the use_device_addr list
+ // whilst maintaining the ordering of useDeviceLocs, useDeviceSyms and
+ // useDeviceTypes to use_device_ptr/use_device_addr input for BlockArg
+ // ordering.
+ // TODO: Perhaps create a user provideable compiler option that will
+ // re-introduce a hard-error rather than a warning in these cases.
+ promoteNonCPtrUseDevicePtrArgsToUseDeviceAddr(clauseOps, useDeviceTypes,
+ useDeviceLocs, useDeviceSyms);
+}
+
+static void genTargetEnterExitUpdateDataClauses(
+ Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::StatementContext &stmtCtx,
+ const Fortran::parser::OmpClauseList &clauses, mlir::Location loc,
+ llvm::omp::Directive directive,
+ mlir::omp::TargetEnterExitUpdateDataClauseOps &clauseOps) {
+ ClauseProcessor cp(converter, semaCtx, clauses);
+ cp.processDepend(clauseOps);
+ cp.processDevice(stmtCtx, clauseOps);
+ cp.processIf(directive, clauseOps);
+ cp.processNowait(clauseOps);
+
+ if (directive == llvm::omp::Directive::OMPD_target_update) {
+ cp.processMotionClauses<clause::To>(stmtCtx, clauseOps);
+ cp.processMotionClauses<clause::From>(stmtCtx, clauseOps);
+ } else {
+ cp.processMap(loc, stmtCtx, clauseOps);
+ }
+}
+
+static void genTaskClauses(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::StatementContext &stmtCtx,
+ const Fortran::parser::OmpClauseList &clauses,
+ mlir::Location loc,
+ mlir::omp::TaskClauseOps &clauseOps) {
+ ClauseProcessor cp(converter, semaCtx, clauses);
cp.processAllocate(clauseOps);
cp.processDefault();
+ cp.processDepend(clauseOps);
cp.processFinal(stmtCtx, clauseOps);
- cp.processUntied(clauseOps);
+ cp.processIf(llvm::omp::Directive::OMPD_task, clauseOps);
cp.processMergeable(clauseOps);
cp.processPriority(stmtCtx, clauseOps);
- cp.processDepend(clauseOps);
+ cp.processUntied(clauseOps);
// TODO Support delayed privatization.
- cp.processTODO<clause::InReduction, clause::Detach, clause::Affinity>(
- currentLocation, llvm::omp::Directive::OMPD_task);
+ cp.processTODO<clause::Affinity, clause::Detach, clause::InReduction>(
+ loc, llvm::omp::Directive::OMPD_task);
+}
- return genOpWithBody<mlir::omp::TaskOp>(
- OpWithBodyGenInfo(converter, semaCtx, currentLocation, eval)
- .setGenNested(genNested)
- .setClauses(&clauseList),
- clauseOps);
+static void genTaskgroupClauses(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ const Fortran::parser::OmpClauseList &clauses,
+ mlir::Location loc,
+ mlir::omp::TaskgroupClauseOps &clauseOps) {
+ ClauseProcessor cp(converter, semaCtx, clauses);
+ cp.processAllocate(clauseOps);
+ cp.processTODO<clause::TaskReduction>(loc,
+ llvm::omp::Directive::OMPD_taskgroup);
}
-static mlir::omp::TaskgroupOp
-genTaskgroupOp(Fortran::lower::AbstractConverter &converter,
- Fortran::semantics::SemanticsContext &semaCtx,
- Fortran::lower::pft::Evaluation &eval, bool genNested,
- mlir::Location currentLocation,
- const Fortran::parser::OmpClauseList &clauseList) {
- mlir::omp::TaskgroupClauseOps clauseOps;
+static void genTaskwaitClauses(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ const Fortran::parser::OmpClauseList &clauses,
+ mlir::Location loc,
+ mlir::omp::TaskwaitClauseOps &clauseOps) {
+ ClauseProcessor cp(converter, semaCtx, clauses);
+ cp.processTODO<clause::Depend, clause::Nowait>(
+ loc, llvm::omp::Directive::OMPD_taskwait);
+}
- ClauseProcessor cp(converter, semaCtx, clauseList);
+static void genTeamsClauses(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::StatementContext &stmtCtx,
+ const Fortran::parser::OmpClauseList &clauses,
+ mlir::Location loc,
+ mlir::omp::TeamsClauseOps &clauseOps) {
+ ClauseProcessor cp(converter, semaCtx, clauses);
cp.processAllocate(clauseOps);
- cp.processTODO<clause::TaskReduction>(currentLocation,
- llvm::omp::Directive::OMPD_taskgroup);
+ cp.processDefault();
+ cp.processIf(llvm::omp::Directive::OMPD_teams, clauseOps);
+ cp.processNumTeams(stmtCtx, clauseOps);
+ cp.processThreadLimit(stmtCtx, clauseOps);
+ // TODO Support delayed privatization.
- return genOpWithBody<mlir::omp::TaskgroupOp>(
- OpWithBodyGenInfo(converter, semaCtx, currentLocation, eval)
- .setGenNested(genNested)
- .setClauses(&clauseList),
- clauseOps);
+ cp.processTODO<clause::Reduction>(loc, llvm::omp::Directive::OMPD_teams);
}
-// This helper function implements the functionality of "promoting"
-// non-CPTR arguments of use_device_ptr to use_device_addr
-// arguments (automagic conversion of use_device_ptr ->
-// use_device_addr in these cases). The way we do so currently is
-// through the shuffling of operands from the devicePtrOperands to
-// deviceAddrOperands where neccesary and re-organizing the types,
-// locations and symbols to maintain the correct ordering of ptr/addr
-// input -> BlockArg.
-//
-// This effectively implements some deprecated OpenMP functionality
-// that some legacy applications unfortunately depend on
-// (deprecated in specification version 5.2):
-//
-// "If a list item in a use_device_ptr clause is not of type C_PTR,
-// the behavior is as if the list item appeared in a use_device_addr
-// clause. Support for such list items in a use_device_ptr clause
-// is deprecated."
-static void promoteNonCPtrUseDevicePtrArgsToUseDeviceAddr(
- mlir::omp::UseDeviceClauseOps &clauseOps,
- llvm::SmallVectorImpl<mlir::Type> &useDeviceTypes,
- llvm::SmallVectorImpl<mlir::Location> &useDeviceLocs,
- llvm::SmallVectorImpl<const Fortran::semantics::Symbol *>
- &useDeviceSymbols) {
- auto moveElementToBack = [](size_t idx, auto &vector) {
- auto *iter = std::next(vector.begin(), idx);
- vector.push_back(*iter);
- vector.erase(iter);
- };
+static void genWsloopClauses(
+ Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::StatementContext &stmtCtx,
+ Fortran::lower::pft::Evaluation &eval,
+ const Fortran::parser::OmpClauseList &beginClauses,
+ const Fortran::parser::OmpClauseList *endClauses, mlir::Location loc,
+ mlir::omp::WsloopClauseOps &clauseOps,
+ llvm::SmallVectorImpl<const Fortran::semantics::Symbol *> &iv,
+ llvm::SmallVectorImpl<mlir::Type> &reductionTypes,
+ llvm::SmallVectorImpl<const Fortran::semantics::Symbol *> &reductionSyms) {
+ fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
+ ClauseProcessor bcp(converter, semaCtx, beginClauses);
+ bcp.processCollapse(loc, eval, clauseOps, iv);
+ bcp.processOrdered(clauseOps);
+ bcp.processReduction(loc, clauseOps, &reductionTypes, &reductionSyms);
+ bcp.processSchedule(stmtCtx, clauseOps);
+ clauseOps.loopInclusiveAttr = firOpBuilder.getUnitAttr();
+ // TODO Support delayed privatization.
- // Iterate over our use_device_ptr list and shift all non-cptr arguments into
- // use_device_addr.
- for (auto *it = clauseOps.useDevicePtrVars.begin();
- it != clauseOps.useDevicePtrVars.end();) {
- if (!fir::isa_builtin_cptr_type(fir::unwrapRefType(it->getType()))) {
- clauseOps.useDeviceAddrVars.push_back(*it);
- // We have to shuffle the symbols around as well, to maintain
- // the correct Input -> BlockArg for use_device_ptr/use_device_addr.
- // NOTE: However, as map's do not seem to be included currently
- // this isn't as pertinent, but we must try to maintain for
- // future alterations. I believe the reason they are not currently
- // is that the BlockArg assign/lowering needs to be extended
- // to a greater set of types.
- auto idx = std::distance(clauseOps.useDevicePtrVars.begin(), it);
- moveElementToBack(idx, useDeviceTypes);
- moveElementToBack(idx, useDeviceLocs);
- moveElementToBack(idx, useDeviceSymbols);
- it = clauseOps.useDevicePtrVars.erase(it);
- continue;
+ if (ReductionProcessor::doReductionByRef(clauseOps.reductionVars))
+ clauseOps.reductionByRefAttr = firOpBuilder.getUnitAttr();
+
+ if (endClauses) {
+ ClauseProcessor ecp(converter, semaCtx, *endClauses);
+ ecp.processNowait(clauseOps);
+ }
+
+ bcp.processTODO<clause::Allocate, clause::Linear, clause::Order>(
+ loc, llvm::omp::Directive::OMPD_do);
+}
+
+//===----------------------------------------------------------------------===//
+// Code generation functions for leaf constructs
+//===----------------------------------------------------------------------===//
+
+static mlir::omp::BarrierOp
+genBarrierOp(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval, mlir::Location loc) {
+ return converter.getFirOpBuilder().create<mlir::omp::BarrierOp>(loc);
+}
+
+static mlir::omp::CriticalOp
+genCriticalOp(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval, bool genNested,
+ mlir::Location loc,
+ const Fortran::parser::OmpClauseList &clauseList,
+ const std::optional<Fortran::parser::Name> &name) {
+ fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
+ mlir::FlatSymbolRefAttr nameAttr;
+
+ if (name) {
+ std::string nameStr = name->ToString();
+ mlir::ModuleOp mod = firOpBuilder.getModule();
+ auto global = mod.lookupSymbol<mlir::omp::CriticalDeclareOp>(nameStr);
+ if (!global) {
+ mlir::omp::CriticalClauseOps clauseOps;
+ genCriticalDeclareClauses(converter, semaCtx, clauseList, loc, clauseOps,
+ nameStr);
+
+ mlir::OpBuilder modBuilder(mod.getBodyRegion());
+ global = modBuilder.create<mlir::omp::CriticalDeclareOp>(loc, clauseOps);
}
- ++it;
+ nameAttr = mlir::FlatSymbolRefAttr::get(firOpBuilder.getContext(),
+ global.getSymName());
}
+
+ return genOpWithBody<mlir::omp::CriticalOp>(
+ OpWithBodyGenInfo(converter, semaCtx, loc, eval).setGenNested(genNested),
+ nameAttr);
}
-static mlir::omp::TargetDataOp
-genTargetDataOp(Fortran::lower::AbstractConverter &converter,
+static mlir::omp::DistributeOp
+genDistributeOp(Fortran::lower::AbstractConverter &converter,
Fortran::semantics::SemanticsContext &semaCtx,
Fortran::lower::pft::Evaluation &eval, bool genNested,
- mlir::Location currentLocation,
+ mlir::Location loc,
const Fortran::parser::OmpClauseList &clauseList) {
- Fortran::lower::StatementContext stmtCtx;
- mlir::omp::TargetDataClauseOps clauseOps;
- llvm::SmallVector<mlir::Type> useDeviceTypes;
- llvm::SmallVector<mlir::Location> useDeviceLocs;
- llvm::SmallVector<const Fortran::semantics::Symbol *> useDeviceSyms;
+ TODO(loc, "Distribute construct");
+ return nullptr;
+}
- ClauseProcessor cp(converter, semaCtx, clauseList);
- cp.processIf(llvm::omp::Directive::OMPD_target_data, clauseOps);
- cp.processDevice(stmtCtx, clauseOps);
- cp.processUseDevicePtr(clauseOps, useDeviceTypes, useDeviceLocs,
- useDeviceSyms);
- cp.processUseDeviceAddr(clauseOps, useDeviceTypes, useDeviceLocs,
- useDeviceSyms);
+static mlir::omp::FlushOp
+genFlushOp(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval, mlir::Location loc,
+ const std::optional<Fortran::parser::OmpObjectList> &objectList,
+ const std::optional<std::list<Fortran::parser::OmpMemoryOrderClause>>
+ &clauseList) {
+ llvm::SmallVector<mlir::Value> operandRange;
+ genFlushClauses(converter, semaCtx, objectList, clauseList, loc,
+ operandRange);
+
+ return converter.getFirOpBuilder().create<mlir::omp::FlushOp>(
+ converter.getCurrentLocation(), operandRange);
+}
- // This function implements the deprecated functionality of use_device_ptr
- // that allows users to provide non-CPTR arguments to it with the caveat
- // that the compiler will treat them as use_device_addr. A lot of legacy
- // code may still depend on this functionality, so we should support it
- // in some manner. We do so currently by simply shifting non-cptr operands
- // from the use_device_ptr list into the front of the use_device_addr list
- // whilst maintaining the ordering of useDeviceLocs, useDeviceSymbols and
- // useDeviceTypes to use_device_ptr/use_device_addr input for BlockArg
- // ordering.
- // TODO: Perhaps create a user provideable compiler option that will
- // re-introduce a hard-error rather than a warning in these cases.
- promoteNonCPtrUseDevicePtrArgsToUseDeviceAddr(clauseOps, useDeviceTypes,
- useDeviceLocs, useDeviceSyms);
- cp.processMap(currentLocation, llvm::omp::Directive::OMPD_target_data,
- stmtCtx, clauseOps);
+static mlir::omp::MasterOp
+genMasterOp(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval, bool genNested,
+ mlir::Location loc) {
+ return genOpWithBody<mlir::omp::MasterOp>(
+ OpWithBodyGenInfo(converter, semaCtx, loc, eval).setGenNested(genNested),
+ /*resultTypes=*/mlir::TypeRange());
+}
+
+static mlir::omp::OrderedOp
+genOrderedOp(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval, mlir::Location loc,
+ const Fortran::parser::OmpClauseList &clauseList) {
+ TODO(loc, "OMPD_ordered");
+ return nullptr;
+}
- auto dataOp = converter.getFirOpBuilder().create<mlir::omp::TargetDataOp>(
- currentLocation, clauseOps);
+static mlir::omp::OrderedRegionOp
+genOrderedRegionOp(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval, bool genNested,
+ mlir::Location loc,
+ const Fortran::parser::OmpClauseList &clauseList) {
+ mlir::omp::OrderedRegionClauseOps clauseOps;
+ genOrderedRegionClauses(converter, semaCtx, clauseList, loc, clauseOps);
- genBodyOfTargetDataOp(converter, semaCtx, eval, genNested, dataOp,
- useDeviceTypes, useDeviceLocs, useDeviceSyms,
- currentLocation);
- return dataOp;
+ return genOpWithBody<mlir::omp::OrderedRegionOp>(
+ OpWithBodyGenInfo(converter, semaCtx, loc, eval).setGenNested(genNested),
+ clauseOps);
}
-template <typename OpTy>
-static OpTy genTargetEnterExitDataUpdateOp(
- Fortran::lower::AbstractConverter &converter,
- Fortran::semantics::SemanticsContext &semaCtx,
- mlir::Location currentLocation,
- const Fortran::parser::OmpClauseList &clauseList) {
+static mlir::omp::ParallelOp
+genParallelOp(Fortran::lower::AbstractConverter &converter,
+ Fortran::lower::SymMap &symTable,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval, bool genNested,
+ mlir::Location loc,
+ const Fortran::parser::OmpClauseList &clauseList,
+ bool outerCombined = false) {
fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
Fortran::lower::StatementContext stmtCtx;
- mlir::omp::TargetEnterExitUpdateDataClauseOps clauseOps;
+ mlir::omp::ParallelClauseOps clauseOps;
+ llvm::SmallVector<const Fortran::semantics::Symbol *> privateSyms;
+ llvm::SmallVector<mlir::Type> reductionTypes;
+ llvm::SmallVector<const Fortran::semantics::Symbol *> reductionSyms;
+ genParallelClauses(converter, semaCtx, stmtCtx, clauseList, loc,
+ /*processReduction=*/!outerCombined, clauseOps,
+ reductionTypes, reductionSyms);
- // GCC 9.3.0 emits a (probably) bogus warning about an unused variable.
- [[maybe_unused]] llvm::omp::Directive directive;
- if constexpr (std::is_same_v<OpTy, mlir::omp::TargetEnterDataOp>) {
- directive = llvm::omp::Directive::OMPD_target_enter_data;
- } else if constexpr (std::is_same_v<OpTy, mlir::omp::TargetExitDataOp>) {
- directive = llvm::omp::Directive::OMPD_target_exit_data;
- } else if constexpr (std::is_same_v<OpTy, mlir::omp::TargetUpdateOp>) {
- directive = llvm::omp::Directive::OMPD_target_update;
- } else {
- return nullptr;
- }
+ auto reductionCallback = [&](mlir::Operation *op) {
+ genReductionVars(op, converter, loc, reductionSyms, reductionTypes);
+ return reductionSyms;
+ };
- ClauseProcessor cp(converter, semaCtx, clauseList);
- cp.processIf(directive, clauseOps);
- cp.processDevice(stmtCtx, clauseOps);
- cp.processDepend(clauseOps);
- cp.processNowait(clauseOps);
+ OpWithBodyGenInfo genInfo =
+ OpWithBodyGenInfo(converter, semaCtx, loc, eval)
+ .setGenNested(genNested)
+ .setOuterCombined(outerCombined)
+ .setClauses(&clauseList)
+ .setReductions(&reductionSyms, &reductionTypes)
+ .setGenRegionEntryCb(reductionCallback);
- if constexpr (std::is_same_v<OpTy, mlir::omp::TargetUpdateOp>) {
- cp.processMotionClauses<clause::To>(stmtCtx, clauseOps);
- cp.processMotionClauses<clause::From>(stmtCtx, clauseOps);
- } else {
- cp.processMap(currentLocation, directive, stmtCtx, clauseOps);
- }
+ if (!enableDelayedPrivatization)
+ return genOpWithBody<mlir::omp::ParallelOp>(genInfo, clauseOps);
- return firOpBuilder.create<OpTy>(currentLocation, clauseOps);
-}
+ bool privatize = !outerCombined;
+ DataSharingProcessor dsp(converter, semaCtx, clauseList, eval,
+ /*useDelayedPrivatization=*/true, &symTable);
-// This functions creates a block for the body of the targetOp's region. It adds
-// all the symbols present in mapSymbols as block arguments to this block.
-static void
-genBodyOfTargetOp(Fortran::lower::AbstractConverter &converter,
- Fortran::semantics::SemanticsContext &semaCtx,
- Fortran::lower::pft::Evaluation &eval, bool genNested,
- mlir::omp::TargetOp &targetOp,
- llvm::ArrayRef<const Fortran::semantics::Symbol *> mapSyms,
- llvm::ArrayRef<mlir::Location> mapSymLocs,
- llvm::ArrayRef<mlir::Type> mapSymTypes,
- const mlir::Location ¤tLocation) {
- assert(mapSymTypes.size() == mapSymLocs.size());
+ if (privatize)
+ dsp.processStep1(&clauseOps, &privateSyms);
- fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
- mlir::Region ®ion = targetOp.getRegion();
+ auto genRegionEntryCB = [&](mlir::Operation *op) {
+ auto parallelOp = llvm::cast<mlir::omp::ParallelOp>(op);
- auto *regionBlock =
- firOpBuilder.createBlock(®ion, {}, mapSymTypes, mapSymLocs);
+ llvm::SmallVector<mlir::Location> reductionLocs(
+ clauseOps.reductionVars.size(), loc);
- // Clones the `bounds` placing them inside the target region and returns them.
- auto cloneBound = [&](mlir::Value bound) {
- if (mlir::isMemoryEffectFree(bound.getDefiningOp())) {
- mlir::Operation *clonedOp = bound.getDefiningOp()->clone();
- regionBlock->push_back(clonedOp);
- return clonedOp->getResult(0);
+ mlir::OperandRange privateVars = parallelOp.getPrivateVars();
+ mlir::Region ®ion = parallelOp.getRegion();
+
+ llvm::SmallVector<mlir::Type> privateVarTypes = reductionTypes;
+ privateVarTypes.reserve(privateVarTypes.size() + privateVars.size());
+ llvm::transform(privateVars, std::back_inserter(privateVarTypes),
+ [](mlir::Value v) { return v.getType(); });
+
+ llvm::SmallVector<mlir::Location> privateVarLocs = reductionLocs;
+ privateVarLocs.reserve(privateVarLocs.size() + privateVars.size());
+ llvm::transform(privateVars, std::back_inserter(privateVarLocs),
+ [](mlir::Value v) { return v.getLoc(); });
+
+ firOpBuilder.createBlock(®ion, /*insertPt=*/{}, privateVarTypes,
+ privateVarLocs);
+
+ llvm::SmallVector<const Fortran::semantics::Symbol *> allSymbols =
+ reductionSyms;
+ allSymbols.append(privateSyms);
+ for (auto [arg, prv] : llvm::zip_equal(allSymbols, region.getArguments())) {
+ converter.bindSymbol(*arg, prv);
}
- TODO(converter.getCurrentLocation(),
- "target map clause operand unsupported bound type");
- };
- auto cloneBounds = [cloneBound](llvm::ArrayRef<mlir::Value> bounds) {
- llvm::SmallVector<mlir::Value> clonedBounds;
- for (mlir::Value bound : bounds)
- clonedBounds.emplace_back(cloneBound(bound));
- return clonedBounds;
+ return allSymbols;
};
- // Bind the symbols to their corresponding block arguments.
- for (auto [argIndex, argSymbol] : llvm::enumerate(mapSyms)) {
- const mlir::BlockArgument &arg = region.getArgument(argIndex);
- // Avoid capture of a reference to a structured binding.
- const Fortran::semantics::Symbol *sym = argSymbol;
- // Structure component symbols don't have bindings.
- if (sym->owner().IsDerivedType())
- continue;
- fir::ExtendedValue extVal = converter.getSymbolExtendedValue(*sym);
- extVal.match(
- [&](const fir::BoxValue &v) {
- converter.bindSymbol(*sym,
- fir::BoxValue(arg, cloneBounds(v.getLBounds()),
- v.getExplicitParameters(),
- v.getExplicitExtents()));
- },
- [&](const fir::MutableBoxValue &v) {
- converter.bindSymbol(
- *sym, fir::MutableBoxValue(arg, cloneBounds(v.getLBounds()),
- v.getMutableProperties()));
- },
- [&](const fir::ArrayBoxValue &v) {
- converter.bindSymbol(
- *sym, fir::ArrayBoxValue(arg, cloneBounds(v.getExtents()),
- cloneBounds(v.getLBounds()),
- v.getSourceBox()));
- },
- [&](const fir::CharArrayBoxValue &v) {
- converter.bindSymbol(
- *sym, fir::CharArrayBoxValue(arg, cloneBound(v.getLen()),
- cloneBounds(v.getExtents()),
- cloneBounds(v.getLBounds())));
- },
- [&](const fir::CharBoxValue &v) {
- converter.bindSymbol(*sym,
- fir::CharBoxValue(arg, cloneBound(v.getLen())));
- },
- [&](const fir::UnboxedValue &v) { converter.bindSymbol(*sym, arg); },
- [&](const auto &) {
- TODO(converter.getCurrentLocation(),
- "target map clause operand unsupported type");
- });
- }
+ // TODO Merge with the reduction CB.
+ genInfo.setGenRegionEntryCb(genRegionEntryCB).setDataSharingProcessor(&dsp);
+ return genOpWithBody<mlir::omp::ParallelOp>(genInfo, clauseOps);
+}
- // Check if cloning the bounds introduced any dependency on the outer region.
- // If so, then either clone them as well if they are MemoryEffectFree, or else
- // copy them to a new temporary and add them to the map and block_argument
- // lists and replace their uses with the new temporary.
- llvm::SetVector<mlir::Value> valuesDefinedAbove;
- mlir::getUsedValuesDefinedAbove(region, valuesDefinedAbove);
- while (!valuesDefinedAbove.empty()) {
- for (mlir::Value val : valuesDefinedAbove) {
- mlir::Operation *valOp = val.getDefiningOp();
- if (mlir::isMemoryEffectFree(valOp)) {
- mlir::Operation *clonedOp = valOp->clone();
- regionBlock->push_front(clonedOp);
- val.replaceUsesWithIf(
- clonedOp->getResult(0), [regionBlock](mlir::OpOperand &use) {
- return use.getOwner()->getBlock() == regionBlock;
- });
- } else {
- auto savedIP = firOpBuilder.getInsertionPoint();
- firOpBuilder.setInsertionPointAfter(valOp);
- auto copyVal =
- firOpBuilder.createTemporary(val.getLoc(), val.getType());
- firOpBuilder.createStoreWithConvert(copyVal.getLoc(), val, copyVal);
+static mlir::omp::SectionOp
+genSectionOp(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval, bool genNested,
+ mlir::Location loc,
+ const Fortran::parser::OmpClauseList &clauseList) {
+ // Currently only private/firstprivate clause is handled, and
+ // all privatization is done within `omp.section` operations.
+ return genOpWithBody<mlir::omp::SectionOp>(
+ OpWithBodyGenInfo(converter, semaCtx, loc, eval)
+ .setGenNested(genNested)
+ .setClauses(&clauseList));
+}
- llvm::SmallVector<mlir::Value> bounds;
- std::stringstream name;
- firOpBuilder.setInsertionPoint(targetOp);
- mlir::Value mapOp = createMapInfoOp(
- firOpBuilder, copyVal.getLoc(), copyVal, mlir::Value{}, name.str(),
- bounds, llvm::SmallVector<mlir::Value>{},
- static_cast<
- std::underlying_type_t<llvm::omp::OpenMPOffloadMappingFlags>>(
- llvm::omp::OpenMPOffloadMappingFlags::OMP_MAP_IMPLICIT),
- mlir::omp::VariableCaptureKind::ByCopy, copyVal.getType());
- targetOp.getMapOperandsMutable().append(mapOp);
- mlir::Value clonedValArg =
- region.addArgument(copyVal.getType(), copyVal.getLoc());
- firOpBuilder.setInsertionPointToStart(regionBlock);
- auto loadOp = firOpBuilder.create<fir::LoadOp>(clonedValArg.getLoc(),
- clonedValArg);
- val.replaceUsesWithIf(
- loadOp->getResult(0), [regionBlock](mlir::OpOperand &use) {
- return use.getOwner()->getBlock() == regionBlock;
- });
- firOpBuilder.setInsertionPoint(regionBlock, savedIP);
- }
- }
- valuesDefinedAbove.clear();
- mlir::getUsedValuesDefinedAbove(region, valuesDefinedAbove);
- }
+static mlir::omp::SectionsOp
+genSectionsOp(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval, mlir::Location loc,
+ const mlir::omp::SectionsClauseOps &clauseOps) {
+ return genOpWithBody<mlir::omp::SectionsOp>(
+ OpWithBodyGenInfo(converter, semaCtx, loc, eval).setGenNested(false),
+ clauseOps);
+}
- // Insert dummy instruction to remember the insertion position. The
- // marker will be deleted since there are not uses.
- // In the HLFIR flow there are hlfir.declares inserted above while
- // setting block arguments.
- mlir::Value undefMarker = firOpBuilder.create<fir::UndefOp>(
- targetOp.getOperation()->getLoc(), firOpBuilder.getIndexType());
+static mlir::omp::SimdLoopOp
+genSimdLoopOp(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval, mlir::Location loc,
+ const Fortran::parser::OmpClauseList &clauseList) {
+ DataSharingProcessor dsp(converter, semaCtx, clauseList, eval);
+ dsp.processStep1();
- // Create blocks for unstructured regions. This has to be done since
- // blocks are initially allocated with the function as the parent region.
- if (eval.lowerAsUnstructured()) {
- Fortran::lower::createEmptyRegionBlocks<mlir::omp::TerminatorOp,
- mlir::omp::YieldOp>(
- firOpBuilder, eval.getNestedEvaluations());
- }
+ Fortran::lower::StatementContext stmtCtx;
+ mlir::omp::SimdLoopClauseOps clauseOps;
+ llvm::SmallVector<const Fortran::semantics::Symbol *> iv;
+ genSimdLoopClauses(converter, semaCtx, stmtCtx, eval, clauseList, loc,
+ clauseOps, iv);
- firOpBuilder.create<mlir::omp::TerminatorOp>(currentLocation);
+ auto *nestedEval =
+ getCollapsedLoopEval(eval, Fortran::lower::getCollapseValue(clauseList));
+
+ auto ivCallback = [&](mlir::Operation *op) {
+ return genLoopVars(op, converter, loc, iv);
+ };
+
+ return genOpWithBody<mlir::omp::SimdLoopOp>(
+ OpWithBodyGenInfo(converter, semaCtx, loc, *nestedEval)
+ .setClauses(&clauseList)
+ .setDataSharingProcessor(&dsp)
+ .setGenRegionEntryCb(ivCallback),
+ clauseOps);
+}
+
+static mlir::omp::SingleOp
+genSingleOp(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval, bool genNested,
+ mlir::Location loc,
+ const Fortran::parser::OmpClauseList &beginClauseList,
+ const Fortran::parser::OmpClauseList &endClauseList) {
+ mlir::omp::SingleClauseOps clauseOps;
+ genSingleClauses(converter, semaCtx, beginClauseList, endClauseList, loc,
+ clauseOps);
- // Create the insertion point after the marker.
- firOpBuilder.setInsertionPointAfter(undefMarker.getDefiningOp());
- if (genNested)
- genNestedEvaluations(converter, eval);
+ return genOpWithBody<mlir::omp::SingleOp>(
+ OpWithBodyGenInfo(converter, semaCtx, loc, eval)
+ .setGenNested(genNested)
+ .setClauses(&beginClauseList),
+ clauseOps);
}
static mlir::omp::TargetOp
genTargetOp(Fortran::lower::AbstractConverter &converter,
Fortran::semantics::SemanticsContext &semaCtx,
Fortran::lower::pft::Evaluation &eval, bool genNested,
- mlir::Location currentLocation,
+ mlir::Location loc,
const Fortran::parser::OmpClauseList &clauseList,
- llvm::omp::Directive directive, bool outerCombined = false) {
+ bool outerCombined = false) {
+ fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
Fortran::lower::StatementContext stmtCtx;
+
+ bool processHostOnlyClauses =
+ !llvm::cast<mlir::omp::OffloadModuleInterface>(*converter.getModuleOp())
+ .getIsTargetDevice();
+
mlir::omp::TargetClauseOps clauseOps;
llvm::SmallVector<const Fortran::semantics::Symbol *> mapSyms;
llvm::SmallVector<mlir::Location> mapSymLocs;
llvm::SmallVector<mlir::Type> mapSymTypes;
-
- ClauseProcessor cp(converter, semaCtx, clauseList);
- cp.processIf(llvm::omp::Directive::OMPD_target, clauseOps);
- cp.processDevice(stmtCtx, clauseOps);
- cp.processThreadLimit(stmtCtx, clauseOps);
- cp.processDepend(clauseOps);
- cp.processNowait(clauseOps);
- cp.processMap(currentLocation, directive, stmtCtx, clauseOps, &mapSyms,
- &mapSymLocs, &mapSymTypes);
- // TODO Support delayed privatization.
-
- cp.processTODO<clause::Private, clause::Firstprivate, clause::IsDevicePtr,
- clause::HasDeviceAddr, clause::Reduction, clause::InReduction,
- clause::Allocate, clause::UsesAllocators, clause::Defaultmap>(
- currentLocation, llvm::omp::Directive::OMPD_target);
+ genTargetClauses(converter, semaCtx, stmtCtx, clauseList, loc,
+ processHostOnlyClauses, /*processReduction=*/outerCombined,
+ clauseOps, mapSyms, mapSymLocs, mapSymTypes);
// 5.8.1 Implicit Data-Mapping Attribute Rules
// The following code follows the implicit data-mapping rules to map all the
@@ -1131,338 +1584,145 @@ genTargetOp(Fortran::lower::AbstractConverter &converter,
};
Fortran::lower::pft::visitAllSymbols(eval, captureImplicitMap);
- auto targetOp = converter.getFirOpBuilder().create<mlir::omp::TargetOp>(
- currentLocation, clauseOps);
-
- genBodyOfTargetOp(converter, semaCtx, eval, genNested, targetOp, mapSyms,
- mapSymLocs, mapSymTypes, currentLocation);
-
- return targetOp;
-}
-
-static mlir::omp::TeamsOp
-genTeamsOp(Fortran::lower::AbstractConverter &converter,
- Fortran::semantics::SemanticsContext &semaCtx,
- Fortran::lower::pft::Evaluation &eval, bool genNested,
- mlir::Location currentLocation,
- const Fortran::parser::OmpClauseList &clauseList,
- bool outerCombined = false) {
- Fortran::lower::StatementContext stmtCtx;
- mlir::omp::TeamsClauseOps clauseOps;
-
- ClauseProcessor cp(converter, semaCtx, clauseList);
- cp.processIf(llvm::omp::Directive::OMPD_teams, clauseOps);
- cp.processAllocate(clauseOps);
- cp.processDefault();
- cp.processNumTeams(stmtCtx, clauseOps);
- cp.processThreadLimit(stmtCtx, clauseOps);
- // TODO Support delayed privatization.
-
- cp.processTODO<clause::Reduction>(currentLocation,
- llvm::omp::Directive::OMPD_teams);
-
- return genOpWithBody<mlir::omp::TeamsOp>(
- OpWithBodyGenInfo(converter, semaCtx, currentLocation, eval)
- .setGenNested(genNested)
- .setOuterCombined(outerCombined)
- .setClauses(&clauseList),
- clauseOps);
-}
-
-/// Extract the list of function and variable symbols affected by the given
-/// 'declare target' directive and return the intended device type for them.
-static void getDeclareTargetInfo(
- Fortran::lower::AbstractConverter &converter,
- Fortran::semantics::SemanticsContext &semaCtx,
- Fortran::lower::pft::Evaluation &eval,
- const Fortran::parser::OpenMPDeclareTargetConstruct &declareTargetConstruct,
- mlir::omp::DeclareTargetClauseOps &clauseOps,
- llvm::SmallVectorImpl<DeclareTargetCapturePair> &symbolAndClause) {
- const auto &spec = std::get<Fortran::parser::OmpDeclareTargetSpecifier>(
- declareTargetConstruct.t);
- if (const auto *objectList{
- Fortran::parser::Unwrap<Fortran::parser::OmpObjectList>(spec.u)}) {
- ObjectList objects{makeList(*objectList, semaCtx)};
- // Case: declare target(func, var1, var2)
- gatherFuncAndVarSyms(objects, mlir::omp::DeclareTargetCaptureClause::to,
- symbolAndClause);
- } else if (const auto *clauseList{
- Fortran::parser::Unwrap<Fortran::parser::OmpClauseList>(
- spec.u)}) {
- if (clauseList->v.empty()) {
- // Case: declare target, implicit capture of function
- symbolAndClause.emplace_back(
- mlir::omp::DeclareTargetCaptureClause::to,
- eval.getOwningProcedure()->getSubprogramSymbol());
- }
-
- ClauseProcessor cp(converter, semaCtx, *clauseList);
- cp.processTo(symbolAndClause);
- cp.processEnter(symbolAndClause);
- cp.processLink(symbolAndClause);
- cp.processDeviceType(clauseOps);
- cp.processTODO<clause::Indirect>(converter.getCurrentLocation(),
- llvm::omp::Directive::OMPD_declare_target);
- }
-}
-
-static void collectDeferredDeclareTargets(
- Fortran::lower::AbstractConverter &converter,
- Fortran::semantics::SemanticsContext &semaCtx,
- Fortran::lower::pft::Evaluation &eval,
- const Fortran::parser::OpenMPDeclareTargetConstruct &declareTargetConstruct,
- llvm::SmallVectorImpl<Fortran::lower::OMPDeferredDeclareTargetInfo>
- &deferredDeclareTarget) {
- mlir::omp::DeclareTargetClauseOps clauseOps;
- llvm::SmallVector<DeclareTargetCapturePair> symbolAndClause;
- getDeclareTargetInfo(converter, semaCtx, eval, declareTargetConstruct,
- clauseOps, symbolAndClause);
- // Return the device type only if at least one of the targets for the
- // directive is a function or subroutine
- mlir::ModuleOp mod = converter.getFirOpBuilder().getModule();
-
- for (const DeclareTargetCapturePair &symClause : symbolAndClause) {
- mlir::Operation *op = mod.lookupSymbol(converter.mangleName(
- std::get<const Fortran::semantics::Symbol &>(symClause)));
-
- if (!op) {
- deferredDeclareTarget.push_back({std::get<0>(symClause),
- clauseOps.deviceType,
- std::get<1>(symClause)});
- }
- }
-}
-
-static std::optional<mlir::omp::DeclareTargetDeviceType>
-getDeclareTargetFunctionDevice(
- Fortran::lower::AbstractConverter &converter,
- Fortran::semantics::SemanticsContext &semaCtx,
- Fortran::lower::pft::Evaluation &eval,
- const Fortran::parser::OpenMPDeclareTargetConstruct
- &declareTargetConstruct) {
- mlir::omp::DeclareTargetClauseOps clauseOps;
- llvm::SmallVector<DeclareTargetCapturePair> symbolAndClause;
- getDeclareTargetInfo(converter, semaCtx, eval, declareTargetConstruct,
- clauseOps, symbolAndClause);
-
- // Return the device type only if at least one of the targets for the
- // directive is a function or subroutine
- mlir::ModuleOp mod = converter.getFirOpBuilder().getModule();
- for (const DeclareTargetCapturePair &symClause : symbolAndClause) {
- mlir::Operation *op = mod.lookupSymbol(converter.mangleName(
- std::get<const Fortran::semantics::Symbol &>(symClause)));
-
- if (mlir::isa_and_nonnull<mlir::func::FuncOp>(op))
- return clauseOps.deviceType;
- }
-
- return std::nullopt;
-}
-
-//===----------------------------------------------------------------------===//
-// genOMP() Code generation helper functions
-//===----------------------------------------------------------------------===//
-
-static void
-genOmpSimpleStandalone(Fortran::lower::AbstractConverter &converter,
- Fortran::semantics::SemanticsContext &semaCtx,
- Fortran::lower::pft::Evaluation &eval, bool genNested,
- const Fortran::parser::OpenMPSimpleStandaloneConstruct
- &simpleStandaloneConstruct) {
- const auto &directive =
- std::get<Fortran::parser::OmpSimpleStandaloneDirective>(
- simpleStandaloneConstruct.t);
- fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
- const auto &opClauseList =
- std::get<Fortran::parser::OmpClauseList>(simpleStandaloneConstruct.t);
- mlir::Location currentLocation = converter.genLocation(directive.source);
-
- switch (directive.v) {
- default:
- break;
- case llvm::omp::Directive::OMPD_barrier:
- firOpBuilder.create<mlir::omp::BarrierOp>(currentLocation);
- break;
- case llvm::omp::Directive::OMPD_taskwait: {
- mlir::omp::TaskwaitClauseOps clauseOps;
- ClauseProcessor cp(converter, semaCtx, opClauseList);
- cp.processTODO<clause::Depend, clause::Nowait>(
- currentLocation, llvm::omp::Directive::OMPD_taskwait);
- firOpBuilder.create<mlir::omp::TaskwaitOp>(currentLocation, clauseOps);
- break;
- }
- case llvm::omp::Directive::OMPD_taskyield:
- firOpBuilder.create<mlir::omp::TaskyieldOp>(currentLocation);
- break;
- case llvm::omp::Directive::OMPD_target_data:
- genTargetDataOp(converter, semaCtx, eval, genNested, currentLocation,
- opClauseList);
- break;
- case llvm::omp::Directive::OMPD_target_enter_data:
- genTargetEnterExitDataUpdateOp<mlir::omp::TargetEnterDataOp>(
- converter, semaCtx, currentLocation, opClauseList);
- break;
- case llvm::omp::Directive::OMPD_target_exit_data:
- genTargetEnterExitDataUpdateOp<mlir::omp::TargetExitDataOp>(
- converter, semaCtx, currentLocation, opClauseList);
- break;
- case llvm::omp::Directive::OMPD_target_update:
- genTargetEnterExitDataUpdateOp<mlir::omp::TargetUpdateOp>(
- converter, semaCtx, currentLocation, opClauseList);
- break;
- case llvm::omp::Directive::OMPD_ordered:
- TODO(currentLocation, "OMPD_ordered");
- }
-}
-
-static void
-genOmpFlush(Fortran::lower::AbstractConverter &converter,
- Fortran::semantics::SemanticsContext &semaCtx,
- Fortran::lower::pft::Evaluation &eval,
- const Fortran::parser::OpenMPFlushConstruct &flushConstruct) {
- llvm::SmallVector<mlir::Value, 4> operandRange;
- if (const auto &ompObjectList =
- std::get<std::optional<Fortran::parser::OmpObjectList>>(
- flushConstruct.t))
- genObjectList2(*ompObjectList, converter, operandRange);
- const auto &memOrderClause =
- std::get<std::optional<std::list<Fortran::parser::OmpMemoryOrderClause>>>(
- flushConstruct.t);
- if (memOrderClause && memOrderClause->size() > 0)
- TODO(converter.getCurrentLocation(), "Handle OmpMemoryOrderClause");
- converter.getFirOpBuilder().create<mlir::omp::FlushOp>(
- converter.getCurrentLocation(), operandRange);
-}
-
-static llvm::SmallVector<const Fortran::semantics::Symbol *>
-genLoopVars(mlir::Operation *op, Fortran::lower::AbstractConverter &converter,
- mlir::Location &loc,
- llvm::ArrayRef<const Fortran::semantics::Symbol *> args) {
- fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
- auto ®ion = op->getRegion(0);
-
- std::size_t loopVarTypeSize = 0;
- for (const Fortran::semantics::Symbol *arg : args)
- loopVarTypeSize = std::max(loopVarTypeSize, arg->GetUltimate().size());
- mlir::Type loopVarType = getLoopVarType(converter, loopVarTypeSize);
- llvm::SmallVector<mlir::Type> tiv(args.size(), loopVarType);
- llvm::SmallVector<mlir::Location> locs(args.size(), loc);
- firOpBuilder.createBlock(®ion, {}, tiv, locs);
- // The argument is not currently in memory, so make a temporary for the
- // argument, and store it there, then bind that location to the argument.
- mlir::Operation *storeOp = nullptr;
- for (auto [argIndex, argSymbol] : llvm::enumerate(args)) {
- mlir::Value indexVal = fir::getBase(region.front().getArgument(argIndex));
- storeOp =
- createAndSetPrivatizedLoopVar(converter, loc, indexVal, argSymbol);
- }
- firOpBuilder.setInsertionPointAfter(storeOp);
-
- return llvm::SmallVector<const Fortran::semantics::Symbol *>(args);
-}
-
-static llvm::SmallVector<const Fortran::semantics::Symbol *>
-genLoopAndReductionVars(
- mlir::Operation *op, Fortran::lower::AbstractConverter &converter,
- mlir::Location &loc,
- llvm::ArrayRef<const Fortran::semantics::Symbol *> loopArgs,
- llvm::ArrayRef<const Fortran::semantics::Symbol *> reductionArgs,
- llvm::ArrayRef<mlir::Type> reductionTypes) {
- fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
-
- llvm::SmallVector<mlir::Type> blockArgTypes;
- llvm::SmallVector<mlir::Location> blockArgLocs;
- blockArgTypes.reserve(loopArgs.size() + reductionArgs.size());
- blockArgLocs.reserve(blockArgTypes.size());
- mlir::Block *entryBlock;
-
- if (loopArgs.size()) {
- std::size_t loopVarTypeSize = 0;
- for (const Fortran::semantics::Symbol *arg : loopArgs)
- loopVarTypeSize = std::max(loopVarTypeSize, arg->GetUltimate().size());
- mlir::Type loopVarType = getLoopVarType(converter, loopVarTypeSize);
- std::fill_n(std::back_inserter(blockArgTypes), loopArgs.size(),
- loopVarType);
- std::fill_n(std::back_inserter(blockArgLocs), loopArgs.size(), loc);
- }
- if (reductionArgs.size()) {
- llvm::copy(reductionTypes, std::back_inserter(blockArgTypes));
- std::fill_n(std::back_inserter(blockArgLocs), reductionArgs.size(), loc);
- }
- entryBlock = firOpBuilder.createBlock(&op->getRegion(0), {}, blockArgTypes,
- blockArgLocs);
- // The argument is not currently in memory, so make a temporary for the
- // argument, and store it there, then bind that location to the argument.
- if (loopArgs.size()) {
- mlir::Operation *storeOp = nullptr;
- for (auto [argIndex, argSymbol] : llvm::enumerate(loopArgs)) {
- mlir::Value indexVal =
- fir::getBase(op->getRegion(0).front().getArgument(argIndex));
- storeOp =
- createAndSetPrivatizedLoopVar(converter, loc, indexVal, argSymbol);
- }
- firOpBuilder.setInsertionPointAfter(storeOp);
- }
- // Bind the reduction arguments to their block arguments
- for (auto [arg, prv] : llvm::zip_equal(
- reductionArgs,
- llvm::drop_begin(entryBlock->getArguments(), loopArgs.size()))) {
- converter.bindSymbol(*arg, prv);
- }
-
- return llvm::SmallVector<const Fortran::semantics::Symbol *>(loopArgs);
+ auto targetOp = firOpBuilder.create<mlir::omp::TargetOp>(loc, clauseOps);
+ genBodyOfTargetOp(converter, semaCtx, eval, genNested, targetOp, mapSyms,
+ mapSymLocs, mapSymTypes, loc);
+ return targetOp;
}
-static void
-createSimdLoop(Fortran::lower::AbstractConverter &converter,
- Fortran::semantics::SemanticsContext &semaCtx,
- Fortran::lower::pft::Evaluation &eval,
- llvm::omp::Directive ompDirective,
- const Fortran::parser::OmpClauseList &loopOpClauseList,
- mlir::Location loc) {
+static mlir::omp::TargetDataOp
+genTargetDataOp(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval, bool genNested,
+ mlir::Location loc,
+ const Fortran::parser::OmpClauseList &clauseList) {
+ Fortran::lower::StatementContext stmtCtx;
+ mlir::omp::TargetDataClauseOps clauseOps;
+ llvm::SmallVector<mlir::Type> useDeviceTypes;
+ llvm::SmallVector<mlir::Location> useDeviceLocs;
+ llvm::SmallVector<const Fortran::semantics::Symbol *> useDeviceSyms;
+ genTargetDataClauses(converter, semaCtx, stmtCtx, clauseList, loc, clauseOps,
+ useDeviceTypes, useDeviceLocs, useDeviceSyms);
+
+ auto targetDataOp =
+ converter.getFirOpBuilder().create<mlir::omp::TargetDataOp>(loc,
+ clauseOps);
+
+ genBodyOfTargetDataOp(converter, semaCtx, eval, genNested, targetDataOp,
+ useDeviceTypes, useDeviceLocs, useDeviceSyms, loc);
+ return targetDataOp;
+}
+
+template <typename OpTy>
+static OpTy genTargetEnterExitUpdateDataOp(
+ Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx, mlir::Location loc,
+ const Fortran::parser::OmpClauseList &clauseList) {
fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
- DataSharingProcessor dsp(converter, semaCtx, loopOpClauseList, eval);
- dsp.processStep1();
+ Fortran::lower::StatementContext stmtCtx;
+
+ // GCC 9.3.0 emits a (probably) bogus warning about an unused variable.
+ [[maybe_unused]] llvm::omp::Directive directive;
+ if constexpr (std::is_same_v<OpTy, mlir::omp::TargetEnterDataOp>) {
+ directive = llvm::omp::Directive::OMPD_target_enter_data;
+ } else if constexpr (std::is_same_v<OpTy, mlir::omp::TargetExitDataOp>) {
+ directive = llvm::omp::Directive::OMPD_target_exit_data;
+ } else if constexpr (std::is_same_v<OpTy, mlir::omp::TargetUpdateOp>) {
+ directive = llvm::omp::Directive::OMPD_target_update;
+ } else {
+ llvm_unreachable("Unexpected TARGET DATA construct");
+ }
+
+ mlir::omp::TargetEnterExitUpdateDataClauseOps clauseOps;
+ genTargetEnterExitUpdateDataClauses(converter, semaCtx, stmtCtx, clauseList,
+ loc, directive, clauseOps);
+
+ return firOpBuilder.create<OpTy>(loc, clauseOps);
+}
+static mlir::omp::TaskOp
+genTaskOp(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval, bool genNested,
+ mlir::Location loc,
+ const Fortran::parser::OmpClauseList &clauseList) {
Fortran::lower::StatementContext stmtCtx;
- mlir::omp::SimdLoopClauseOps clauseOps;
- llvm::SmallVector<const Fortran::semantics::Symbol *> iv;
+ mlir::omp::TaskClauseOps clauseOps;
+ genTaskClauses(converter, semaCtx, stmtCtx, clauseList, loc, clauseOps);
- ClauseProcessor cp(converter, semaCtx, loopOpClauseList);
- cp.processCollapse(loc, eval, clauseOps, iv);
- cp.processReduction(loc, clauseOps);
- cp.processIf(llvm::omp::Directive::OMPD_simd, clauseOps);
- cp.processSimdlen(clauseOps);
- cp.processSafelen(clauseOps);
- clauseOps.loopInclusiveAttr = firOpBuilder.getUnitAttr();
- // TODO Support delayed privatization.
+ return genOpWithBody<mlir::omp::TaskOp>(
+ OpWithBodyGenInfo(converter, semaCtx, loc, eval)
+ .setGenNested(genNested)
+ .setClauses(&clauseList),
+ clauseOps);
+}
- cp.processTODO<clause::Aligned, clause::Allocate, clause::Linear,
- clause::Nontemporal, clause::Order>(loc, ompDirective);
+static mlir::omp::TaskgroupOp
+genTaskgroupOp(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval, bool genNested,
+ mlir::Location loc,
+ const Fortran::parser::OmpClauseList &clauseList) {
+ mlir::omp::TaskgroupClauseOps clauseOps;
+ genTaskgroupClauses(converter, semaCtx, clauseList, loc, clauseOps);
- auto *nestedEval = getCollapsedLoopEval(
- eval, Fortran::lower::getCollapseValue(loopOpClauseList));
+ return genOpWithBody<mlir::omp::TaskgroupOp>(
+ OpWithBodyGenInfo(converter, semaCtx, loc, eval)
+ .setGenNested(genNested)
+ .setClauses(&clauseList),
+ clauseOps);
+}
- auto ivCallback = [&](mlir::Operation *op) {
- return genLoopVars(op, converter, loc, iv);
- };
+static mlir::omp::TaskloopOp
+genTaskloopOp(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval, mlir::Location loc,
+ const Fortran::parser::OmpClauseList &clauseList) {
+ TODO(loc, "Taskloop construct");
+}
- genOpWithBody<mlir::omp::SimdLoopOp>(
- OpWithBodyGenInfo(converter, semaCtx, loc, *nestedEval)
- .setClauses(&loopOpClauseList)
- .setDataSharingProcessor(&dsp)
- .setGenRegionEntryCb(ivCallback),
+static mlir::omp::TaskwaitOp
+genTaskwaitOp(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval, mlir::Location loc,
+ const Fortran::parser::OmpClauseList &clauseList) {
+ mlir::omp::TaskwaitClauseOps clauseOps;
+ genTaskwaitClauses(converter, semaCtx, clauseList, loc, clauseOps);
+ return converter.getFirOpBuilder().create<mlir::omp::TaskwaitOp>(loc,
+ clauseOps);
+}
+
+static mlir::omp::TaskyieldOp
+genTaskyieldOp(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval, mlir::Location loc) {
+ return converter.getFirOpBuilder().create<mlir::omp::TaskyieldOp>(loc);
+}
+
+static mlir::omp::TeamsOp
+genTeamsOp(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval, bool genNested,
+ mlir::Location loc, const Fortran::parser::OmpClauseList &clauseList,
+ bool outerCombined = false) {
+ Fortran::lower::StatementContext stmtCtx;
+ mlir::omp::TeamsClauseOps clauseOps;
+ genTeamsClauses(converter, semaCtx, stmtCtx, clauseList, loc, clauseOps);
+
+ return genOpWithBody<mlir::omp::TeamsOp>(
+ OpWithBodyGenInfo(converter, semaCtx, loc, eval)
+ .setGenNested(genNested)
+ .setOuterCombined(outerCombined)
+ .setClauses(&clauseList),
clauseOps);
}
-static void createWsloop(Fortran::lower::AbstractConverter &converter,
- Fortran::semantics::SemanticsContext &semaCtx,
- Fortran::lower::pft::Evaluation &eval,
- llvm::omp::Directive ompDirective,
- const Fortran::parser::OmpClauseList &beginClauseList,
- const Fortran::parser::OmpClauseList *endClauseList,
- mlir::Location loc) {
- fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
+static mlir::omp::WsloopOp
+genWsloopOp(Fortran::lower::AbstractConverter &converter,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval, mlir::Location loc,
+ const Fortran::parser::OmpClauseList &beginClauseList,
+ const Fortran::parser::OmpClauseList *endClauseList) {
DataSharingProcessor dsp(converter, semaCtx, beginClauseList, eval);
dsp.processStep1();
@@ -1471,30 +1731,9 @@ static void createWsloop(Fortran::lower::AbstractConverter &converter,
llvm::SmallVector<const Fortran::semantics::Symbol *> iv;
llvm::SmallVector<mlir::Type> reductionTypes;
llvm::SmallVector<const Fortran::semantics::Symbol *> reductionSyms;
-
- ClauseProcessor cp(converter, semaCtx, beginClauseList);
- cp.processCollapse(loc, eval, clauseOps, iv);
- cp.processSchedule(stmtCtx, clauseOps);
- cp.processReduction(loc, clauseOps, &reductionTypes, &reductionSyms);
- cp.processOrdered(clauseOps);
- clauseOps.loopInclusiveAttr = firOpBuilder.getUnitAttr();
- // TODO Support delayed privatization.
-
- if (ReductionProcessor::doReductionByRef(clauseOps.reductionVars))
- clauseOps.reductionByRefAttr = firOpBuilder.getUnitAttr();
-
- cp.processTODO<clause::Allocate, clause::Linear, clause::Order>(loc,
- ompDirective);
-
- // In FORTRAN `nowait` clause occur at the end of `omp do` directive.
- // i.e
- // !$omp do
- // <...>
- // !$omp end do nowait
- if (endClauseList) {
- ClauseProcessor ecp(converter, semaCtx, *endClauseList);
- ecp.processNowait(clauseOps);
- }
+ genWsloopClauses(converter, semaCtx, stmtCtx, eval, beginClauseList,
+ endClauseList, loc, clauseOps, iv, reductionTypes,
+ reductionSyms);
auto *nestedEval = getCollapsedLoopEval(
eval, Fortran::lower::getCollapseValue(beginClauseList));
@@ -1504,7 +1743,7 @@ static void createWsloop(Fortran::lower::AbstractConverter &converter,
reductionTypes);
};
- genOpWithBody<mlir::omp::WsloopOp>(
+ return genOpWithBody<mlir::omp::WsloopOp>(
OpWithBodyGenInfo(converter, semaCtx, loc, *nestedEval)
.setClauses(&beginClauseList)
.setDataSharingProcessor(&dsp)
@@ -1513,7 +1752,11 @@ static void createWsloop(Fortran::lower::AbstractConverter &converter,
clauseOps);
}
-static void createSimdWsloop(
+//===----------------------------------------------------------------------===//
+// Code generation functions for composite constructs
+//===----------------------------------------------------------------------===//
+
+static void genCompositeDoSimd(
Fortran::lower::AbstractConverter &converter,
Fortran::semantics::SemanticsContext &semaCtx,
Fortran::lower::pft::Evaluation &eval, llvm::omp::Directive ompDirective,
@@ -1521,7 +1764,7 @@ static void createSimdWsloop(
const Fortran::parser::OmpClauseList *endClauseList, mlir::Location loc) {
ClauseProcessor cp(converter, semaCtx, beginClauseList);
cp.processTODO<clause::Aligned, clause::Allocate, clause::Linear,
- clause::Safelen, clause::Simdlen, clause::Order>(loc,
+ clause::Order, clause::Safelen, clause::Simdlen>(loc,
ompDirective);
// TODO: Add support for vectorization - add vectorization hints inside loop
// body.
@@ -1531,34 +1774,7 @@ static void createSimdWsloop(
// When support for vectorization is enabled, then we need to add handling of
// if clause. Currently if clause can be skipped because we always assume
// SIMD length = 1.
- createWsloop(converter, semaCtx, eval, ompDirective, beginClauseList,
- endClauseList, loc);
-}
-
-static void
-markDeclareTarget(mlir::Operation *op,
- Fortran::lower::AbstractConverter &converter,
- mlir::omp::DeclareTargetCaptureClause captureClause,
- mlir::omp::DeclareTargetDeviceType deviceType) {
- // TODO: Add support for program local variables with declare target applied
- auto declareTargetOp = llvm::dyn_cast<mlir::omp::DeclareTargetInterface>(op);
- if (!declareTargetOp)
- fir::emitFatalError(
- converter.getCurrentLocation(),
- "Attempt to apply declare target on unsupported operation");
-
- // The function or global already has a declare target applied to it, very
- // likely through implicit capture (usage in another declare target
- // function/subroutine). It should be marked as any if it has been assigned
- // both host and nohost, else we skip, as there is no change
- if (declareTargetOp.isDeclareTarget()) {
- if (declareTargetOp.getDeclareTargetDeviceType() != deviceType)
- declareTargetOp.setDeclareTarget(mlir::omp::DeclareTargetDeviceType::any,
- captureClause);
- return;
- }
-
- declareTargetOp.setDeclareTarget(deviceType, captureClause);
+ genWsloopOp(converter, semaCtx, eval, loc, beginClauseList, endClauseList);
}
//===----------------------------------------------------------------------===//
@@ -1653,6 +1869,102 @@ genOMP(Fortran::lower::AbstractConverter &converter,
ompDeclConstruct.u);
}
+//===----------------------------------------------------------------------===//
+// OpenMPStandaloneConstruct visitors
+//===----------------------------------------------------------------------===//
+
+static void genOMP(Fortran::lower::AbstractConverter &converter,
+ Fortran::lower::SymMap &symTable,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval,
+ const Fortran::parser::OpenMPSimpleStandaloneConstruct
+ &simpleStandaloneConstruct) {
+ const auto &directive =
+ std::get<Fortran::parser::OmpSimpleStandaloneDirective>(
+ simpleStandaloneConstruct.t);
+ const auto &clauseList =
+ std::get<Fortran::parser::OmpClauseList>(simpleStandaloneConstruct.t);
+ mlir::Location currentLocation = converter.genLocation(directive.source);
+
+ switch (directive.v) {
+ default:
+ break;
+ case llvm::omp::Directive::OMPD_barrier:
+ genBarrierOp(converter, semaCtx, eval, currentLocation);
+ break;
+ case llvm::omp::Directive::OMPD_taskwait:
+ genTaskwaitOp(converter, semaCtx, eval, currentLocation, clauseList);
+ break;
+ case llvm::omp::Directive::OMPD_taskyield:
+ genTaskyieldOp(converter, semaCtx, eval, currentLocation);
+ break;
+ case llvm::omp::Directive::OMPD_target_data:
+ genTargetDataOp(converter, semaCtx, eval, /*genNested=*/true,
+ currentLocation, clauseList);
+ break;
+ case llvm::omp::Directive::OMPD_target_enter_data:
+ genTargetEnterExitUpdateDataOp<mlir::omp::TargetEnterDataOp>(
+ converter, semaCtx, currentLocation, clauseList);
+ break;
+ case llvm::omp::Directive::OMPD_target_exit_data:
+ genTargetEnterExitUpdateDataOp<mlir::omp::TargetExitDataOp>(
+ converter, semaCtx, currentLocation, clauseList);
+ break;
+ case llvm::omp::Directive::OMPD_target_update:
+ genTargetEnterExitUpdateDataOp<mlir::omp::TargetUpdateOp>(
+ converter, semaCtx, currentLocation, clauseList);
+ break;
+ case llvm::omp::Directive::OMPD_ordered:
+ genOrderedOp(converter, semaCtx, eval, currentLocation, clauseList);
+ break;
+ }
+}
+
+static void
+genOMP(Fortran::lower::AbstractConverter &converter,
+ Fortran::lower::SymMap &symTable,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval,
+ const Fortran::parser::OpenMPFlushConstruct &flushConstruct) {
+ const auto &verbatim = std::get<Fortran::parser::Verbatim>(flushConstruct.t);
+ const auto &objectList =
+ std::get<std::optional<Fortran::parser::OmpObjectList>>(flushConstruct.t);
+ const auto &clauseList =
+ std::get<std::optional<std::list<Fortran::parser::OmpMemoryOrderClause>>>(
+ flushConstruct.t);
+ mlir::Location currentLocation = converter.genLocation(verbatim.source);
+ genFlushOp(converter, semaCtx, eval, currentLocation, objectList, clauseList);
+}
+
+static void
+genOMP(Fortran::lower::AbstractConverter &converter,
+ Fortran::lower::SymMap &symTable,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval,
+ const Fortran::parser::OpenMPCancelConstruct &cancelConstruct) {
+ TODO(converter.getCurrentLocation(), "OpenMPCancelConstruct");
+}
+
+static void genOMP(Fortran::lower::AbstractConverter &converter,
+ Fortran::lower::SymMap &symTable,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval,
+ const Fortran::parser::OpenMPCancellationPointConstruct
+ &cancellationPointConstruct) {
+ TODO(converter.getCurrentLocation(), "OpenMPCancelConstruct");
+}
+
+static void
+genOMP(Fortran::lower::AbstractConverter &converter,
+ Fortran::lower::SymMap &symTable,
+ Fortran::semantics::SemanticsContext &semaCtx,
+ Fortran::lower::pft::Evaluation &eval,
+ const Fortran::parser::OpenMPStandaloneConstruct &standaloneConstruct) {
+ std::visit(
+ [&](auto &&s) { return genOMP(converter, symTable, semaCtx, eval, s); },
+ standaloneConstruct.u);
+}
+
//===----------------------------------------------------------------------===//
// OpenMPConstruct visitors
//===----------------------------------------------------------------------===//
@@ -1782,7 +2094,7 @@ genOMP(Fortran::lower::AbstractConverter &converter,
break;
case llvm::omp::Directive::OMPD_target:
genTargetOp(converter, semaCtx, eval, /*genNested=*/true, currentLocation,
- beginClauseList, directive.v);
+ beginClauseList);
break;
case llvm::omp::Directive::OMPD_target_data:
genTargetDataOp(converter, semaCtx, eval, /*genNested=*/true,
@@ -1798,8 +2110,7 @@ genOMP(Fortran::lower::AbstractConverter &converter,
break;
case llvm::omp::Directive::OMPD_teams:
genTeamsOp(converter, semaCtx, eval, /*genNested=*/true, currentLocation,
- beginClauseList,
- /*outerCombined=*/false);
+ beginClauseList);
break;
case llvm::omp::Directive::OMPD_workshare:
// FIXME: Workshare is not a commonly used OpenMP construct, an
@@ -1821,8 +2132,7 @@ genOMP(Fortran::lower::AbstractConverter &converter,
if ((llvm::omp::allTargetSet & llvm::omp::blockConstructSet)
.test(directive.v)) {
genTargetOp(converter, semaCtx, eval, /*genNested=*/false, currentLocation,
- beginClauseList, directive.v,
- /*outerCombined=*/true);
+ beginClauseList, /*outerCombined=*/true);
combinedDirective = true;
}
if ((llvm::omp::allTeamsSet & llvm::omp::blockConstructSet)
@@ -1859,44 +2169,13 @@ genOMP(Fortran::lower::AbstractConverter &converter,
Fortran::semantics::SemanticsContext &semaCtx,
Fortran::lower::pft::Evaluation &eval,
const Fortran::parser::OpenMPCriticalConstruct &criticalConstruct) {
- fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
- mlir::Location currentLocation = converter.getCurrentLocation();
- std::string name;
- const Fortran::parser::OmpCriticalDirective &cd =
+ const auto &cd =
std::get<Fortran::parser::OmpCriticalDirective>(criticalConstruct.t);
- if (std::get<std::optional<Fortran::parser::Name>>(cd.t).has_value()) {
- name =
- std::get<std::optional<Fortran::parser::Name>>(cd.t).value().ToString();
- }
-
- mlir::omp::CriticalOp criticalOp = [&]() {
- if (name.empty()) {
- return firOpBuilder.create<mlir::omp::CriticalOp>(
- currentLocation, mlir::FlatSymbolRefAttr());
- }
-
- mlir::ModuleOp module = firOpBuilder.getModule();
- mlir::OpBuilder modBuilder(module.getBodyRegion());
- auto global = module.lookupSymbol<mlir::omp::CriticalDeclareOp>(name);
- if (!global) {
- mlir::omp::CriticalClauseOps clauseOps;
- const auto &clauseList = std::get<Fortran::parser::OmpClauseList>(cd.t);
-
- ClauseProcessor cp(converter, semaCtx, clauseList);
- cp.processHint(clauseOps);
- clauseOps.nameAttr =
- mlir::StringAttr::get(firOpBuilder.getContext(), name);
-
- global = modBuilder.create<mlir::omp::CriticalDeclareOp>(currentLocation,
- clauseOps);
- }
-
- return firOpBuilder.create<mlir::omp::CriticalOp>(
- currentLocation, mlir::FlatSymbolRefAttr::get(firOpBuilder.getContext(),
- global.getSymName()));
- }();
- auto genInfo = OpWithBodyGenInfo(converter, semaCtx, currentLocation, eval);
- createBodyOfOp<mlir::omp::CriticalOp>(criticalOp, genInfo);
+ const auto &clauseList = std::get<Fortran::parser::OmpClauseList>(cd.t);
+ const auto &name = std::get<std::optional<Fortran::parser::Name>>(cd.t);
+ mlir::Location currentLocation = converter.getCurrentLocation();
+ genCriticalOp(converter, semaCtx, eval, /*genNested=*/true, currentLocation,
+ clauseList, name);
}
static void
@@ -1915,7 +2194,7 @@ static void genOMP(Fortran::lower::AbstractConverter &converter,
const Fortran::parser::OpenMPLoopConstruct &loopConstruct) {
const auto &beginLoopDirective =
std::get<Fortran::parser::OmpBeginLoopDirective>(loopConstruct.t);
- const auto &loopOpClauseList =
+ const auto &beginClauseList =
std::get<Fortran::parser::OmpClauseList>(beginLoopDirective.t);
mlir::Location currentLocation =
converter.genLocation(beginLoopDirective.source);
@@ -1936,33 +2215,31 @@ static void genOMP(Fortran::lower::AbstractConverter &converter,
bool validDirective = false;
if (llvm::omp::topTaskloopSet.test(ompDirective)) {
validDirective = true;
- TODO(currentLocation, "Taskloop construct");
+ genTaskloopOp(converter, semaCtx, eval, currentLocation, beginClauseList);
} else {
// Create omp.{target, teams, distribute, parallel} nested operations
if ((llvm::omp::allTargetSet & llvm::omp::loopConstructSet)
.test(ompDirective)) {
validDirective = true;
genTargetOp(converter, semaCtx, eval, /*genNested=*/false,
- currentLocation, loopOpClauseList, ompDirective,
- /*outerCombined=*/true);
+ currentLocation, beginClauseList, /*outerCombined=*/true);
}
if ((llvm::omp::allTeamsSet & llvm::omp::loopConstructSet)
.test(ompDirective)) {
validDirective = true;
genTeamsOp(converter, semaCtx, eval, /*genNested=*/false, currentLocation,
- loopOpClauseList,
- /*outerCombined=*/true);
+ beginClauseList, /*outerCombined=*/true);
}
if (llvm::omp::allDistributeSet.test(ompDirective)) {
validDirective = true;
- TODO(currentLocation, "Distribute construct");
+ genDistributeOp(converter, semaCtx, eval, /*genNested=*/false,
+ currentLocation, beginClauseList);
}
if ((llvm::omp::allParallelSet & llvm::omp::loopConstructSet)
.test(ompDirective)) {
validDirective = true;
genParallelOp(converter, symTable, semaCtx, eval, /*genNested=*/false,
- currentLocation, loopOpClauseList,
- /*outerCombined=*/true);
+ currentLocation, beginClauseList, /*outerCombined=*/true);
}
}
if ((llvm::omp::allDoSet | llvm::omp::allSimdSet).test(ompDirective))
@@ -1976,17 +2253,15 @@ static void genOMP(Fortran::lower::AbstractConverter &converter,
if (llvm::omp::allDoSimdSet.test(ompDirective)) {
// 2.9.3.2 Workshare SIMD construct
- createSimdWsloop(converter, semaCtx, eval, ompDirective, loopOpClauseList,
- endClauseList, currentLocation);
-
+ genCompositeDoSimd(converter, semaCtx, eval, ompDirective, beginClauseList,
+ endClauseList, currentLocation);
} else if (llvm::omp::allSimdSet.test(ompDirective)) {
// 2.9.3.1 SIMD construct
- createSimdLoop(converter, semaCtx, eval, ompDirective, loopOpClauseList,
- currentLocation);
- genOpenMPReduction(converter, semaCtx, loopOpClauseList);
+ genSimdLoopOp(converter, semaCtx, eval, currentLocation, beginClauseList);
+ genOpenMPReduction(converter, semaCtx, beginClauseList);
} else {
- createWsloop(converter, semaCtx, eval, ompDirective, loopOpClauseList,
- endClauseList, currentLocation);
+ genWsloopOp(converter, semaCtx, eval, currentLocation, beginClauseList,
+ endClauseList);
}
}
@@ -2006,44 +2281,39 @@ genOMP(Fortran::lower::AbstractConverter &converter,
Fortran::semantics::SemanticsContext &semaCtx,
Fortran::lower::pft::Evaluation &eval,
const Fortran::parser::OpenMPSectionsConstruct §ionsConstruct) {
- mlir::Location currentLocation = converter.getCurrentLocation();
- mlir::omp::SectionsClauseOps clauseOps;
const auto &beginSectionsDirective =
std::get<Fortran::parser::OmpBeginSectionsDirective>(sectionsConstruct.t);
- const auto §ionsClauseList =
+ const auto &beginClauseList =
std::get<Fortran::parser::OmpClauseList>(beginSectionsDirective.t);
// Process clauses before optional omp.parallel, so that new variables are
// allocated outside of the parallel region
- ClauseProcessor cp(converter, semaCtx, sectionsClauseList);
- cp.processSectionsReduction(currentLocation, clauseOps);
- cp.processAllocate(clauseOps);
- // TODO Support delayed privatization.
+ mlir::Location currentLocation = converter.getCurrentLocation();
+ mlir::omp::SectionsClauseOps clauseOps;
+ genSectionsClauses(converter, semaCtx, beginClauseList, currentLocation,
+ /*clausesFromBeginSections=*/true, clauseOps);
+ // Parallel wrapper of PARALLEL SECTIONS construct
llvm::omp::Directive dir =
std::get<Fortran::parser::OmpSectionsDirective>(beginSectionsDirective.t)
.v;
-
- // Parallel wrapper of PARALLEL SECTIONS construct
if (dir == llvm::omp::Directive::OMPD_parallel_sections) {
genParallelOp(converter, symTable, semaCtx, eval,
- /*genNested=*/false, currentLocation, sectionsClauseList,
+ /*genNested=*/false, currentLocation, beginClauseList,
/*outerCombined=*/true);
} else {
const auto &endSectionsDirective =
std::get<Fortran::parser::OmpEndSectionsDirective>(sectionsConstruct.t);
- const auto &endSectionsClauseList =
+ const auto &endClauseList =
std::get<Fortran::parser::OmpClauseList>(endSectionsDirective.t);
- ClauseProcessor(converter, semaCtx, endSectionsClauseList)
- .processNowait(clauseOps);
+ genSectionsClauses(converter, semaCtx, endClauseList, currentLocation,
+ /*clausesFromBeginSections=*/false, clauseOps);
}
- // SECTIONS construct
- genOpWithBody<mlir::omp::SectionsOp>(
- OpWithBodyGenInfo(converter, semaCtx, currentLocation, eval)
- .setGenNested(false),
- clauseOps);
+ // SECTIONS construct.
+ genSectionsOp(converter, semaCtx, eval, currentLocation, clauseOps);
+ // Generate nested SECTION operations recursively.
const auto §ionBlocks =
std::get<Fortran::parser::OmpSectionBlocks>(sectionsConstruct.t);
auto &firOpBuilder = converter.getFirOpBuilder();
@@ -2052,40 +2322,12 @@ genOMP(Fortran::lower::AbstractConverter &converter,
llvm::zip(sectionBlocks.v, eval.getNestedEvaluations())) {
symTable.pushScope();
genSectionOp(converter, semaCtx, neval, /*genNested=*/true, currentLocation,
- sectionsClauseList);
+ beginClauseList);
symTable.popScope();
firOpBuilder.restoreInsertionPoint(ip);
}
}
-static void
-genOMP(Fortran::lower::AbstractConverter &converter,
- Fortran::lower::SymMap &symTable,
- Fortran::semantics::SemanticsContext &semaCtx,
- Fortran::lower::pft::Evaluation &eval,
- const Fortran::parser::OpenMPStandaloneConstruct &standaloneConstruct) {
- std::visit(
- Fortran::common::visitors{
- [&](const Fortran::parser::OpenMPSimpleStandaloneConstruct
- &simpleStandaloneConstruct) {
- genOmpSimpleStandalone(converter, semaCtx, eval,
- /*genNested=*/true,
- simpleStandaloneConstruct);
- },
- [&](const Fortran::parser::OpenMPFlushConstruct &flushConstruct) {
- genOmpFlush(converter, semaCtx, eval, flushConstruct);
- },
- [&](const Fortran::parser::OpenMPCancelConstruct &cancelConstruct) {
- TODO(converter.getCurrentLocation(), "OpenMPCancelConstruct");
- },
- [&](const Fortran::parser::OpenMPCancellationPointConstruct
- &cancellationPointConstruct) {
- TODO(converter.getCurrentLocation(), "OpenMPCancelConstruct");
- },
- },
- standaloneConstruct.u);
-}
-
static void genOMP(Fortran::lower::AbstractConverter &converter,
Fortran::lower::SymMap &symTable,
Fortran::semantics::SemanticsContext &semaCtx,
diff --git a/flang/test/Lower/OpenMP/FIR/target.f90 b/flang/test/Lower/OpenMP/FIR/target.f90
index 821196b83c3b99..d3f2a1c7a15936 100644
--- a/flang/test/Lower/OpenMP/FIR/target.f90
+++ b/flang/test/Lower/OpenMP/FIR/target.f90
@@ -411,8 +411,8 @@ end subroutine omp_target_implicit_bounds
!CHECK-LABEL: func.func @_QPomp_target_thread_limit() {
subroutine omp_target_thread_limit
integer :: a
- !CHECK: %[[VAL_1:.*]] = arith.constant 64 : i32
!CHECK: %[[MAP:.*]] = omp.map.info var_ptr({{.*}}) map_clauses(tofrom) capture(ByRef) -> !fir.ref<i32> {name = "a"}
+ !CHECK: %[[VAL_1:.*]] = arith.constant 64 : i32
!CHECK: omp.target thread_limit(%[[VAL_1]] : i32) map_entries(%[[MAP]] -> %[[ARG_0:.*]] : !fir.ref<i32>) {
!CHECK: ^bb0(%[[ARG_0]]: !fir.ref<i32>):
!$omp target map(tofrom: a) thread_limit(64)
diff --git a/flang/test/Lower/OpenMP/target.f90 b/flang/test/Lower/OpenMP/target.f90
index 6f72b5a34d069a..51b66327dfb24b 100644
--- a/flang/test/Lower/OpenMP/target.f90
+++ b/flang/test/Lower/OpenMP/target.f90
@@ -490,8 +490,8 @@ end subroutine omp_target_implicit_bounds
!CHECK-LABEL: func.func @_QPomp_target_thread_limit() {
subroutine omp_target_thread_limit
integer :: a
- !CHECK: %[[VAL_1:.*]] = arith.constant 64 : i32
!CHECK: %[[MAP:.*]] = omp.map.info var_ptr({{.*}}) map_clauses(tofrom) capture(ByRef) -> !fir.ref<i32> {name = "a"}
+ !CHECK: %[[VAL_1:.*]] = arith.constant 64 : i32
!CHECK: omp.target thread_limit(%[[VAL_1]] : i32) map_entries(%[[MAP]] -> %{{.*}} : !fir.ref<i32>) {
!CHECK: ^bb0(%{{.*}}: !fir.ref<i32>):
!$omp target map(tofrom: a) thread_limit(64)
diff --git a/flang/test/Lower/OpenMP/use-device-ptr-to-use-device-addr.f90 b/flang/test/Lower/OpenMP/use-device-ptr-to-use-device-addr.f90
index 33b5971656010a..d849dd206b9439 100644
--- a/flang/test/Lower/OpenMP/use-device-ptr-to-use-device-addr.f90
+++ b/flang/test/Lower/OpenMP/use-device-ptr-to-use-device-addr.f90
@@ -21,7 +21,7 @@ subroutine only_use_device_ptr
!CHECK: func.func @{{.*}}mix_use_device_ptr_and_addr()
!CHECK: omp.target_data use_device_ptr({{.*}} : !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>>) use_device_addr(%{{.*}}, %{{.*}} : !fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>>, !fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>>) {
-!CHECK: ^bb0(%{{.*}}: !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>>, %{{.*}}: !fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>>, %{{.*}}: !fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>>):
+!CHECK: ^bb0(%{{.*}}: !fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>>, %{{.*}}: !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>>, %{{.*}}: !fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>>):
subroutine mix_use_device_ptr_and_addr
use iso_c_binding
integer, pointer, dimension(:) :: array
@@ -47,7 +47,7 @@ subroutine only_use_device_addr
!CHECK: func.func @{{.*}}mix_use_device_ptr_and_addr_and_map()
!CHECK: omp.target_data map_entries(%{{.*}}, %{{.*}} : !fir.ref<i32>, !fir.ref<i32>) use_device_ptr(%{{.*}} : !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>>) use_device_addr(%{{.*}}, %{{.*}} : !fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>>, !fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>>) {
-!CHECK: ^bb0(%{{.*}}: !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>>, %{{.*}}: !fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>>, %{{.*}}: !fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>>):
+!CHECK: ^bb0(%{{.*}}: !fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>>, %{{.*}}: !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_c_ptr{__address:i64}>>, %{{.*}}: !fir.ref<!fir.box<!fir.ptr<!fir.array<?xi32>>>>):
subroutine mix_use_device_ptr_and_addr_and_map
use iso_c_binding
integer :: i, j
More information about the llvm-branch-commits
mailing list