diff --git a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp index a96884f5680ba..55eda7e3404c1 100644 --- a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp +++ b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp @@ -431,6 +431,19 @@ bool ClauseProcessor::processNumTasks( return false; } +bool ClauseProcessor::processSizes(StatementContext &stmtCtx, + mlir::omp::SizesClauseOps &result) const { + if (auto *clause = findUniqueClause()) { + result.sizes.reserve(clause->v.size()); + for (const ExprTy &vv : clause->v) + result.sizes.push_back(fir::getBase(converter.genExprValue(vv, stmtCtx))); + + return true; + } + + return false; +} + bool ClauseProcessor::processNumTeams( lower::StatementContext &stmtCtx, mlir::omp::NumTeamsClauseOps &result) const { diff --git a/flang/lib/Lower/OpenMP/ClauseProcessor.h b/flang/lib/Lower/OpenMP/ClauseProcessor.h index 324ea3c1047a5..9e352fa574a97 100644 --- a/flang/lib/Lower/OpenMP/ClauseProcessor.h +++ b/flang/lib/Lower/OpenMP/ClauseProcessor.h @@ -66,6 +66,8 @@ class ClauseProcessor { mlir::omp::LoopRelatedClauseOps &loopResult, mlir::omp::CollapseClauseOps &collapseResult, llvm::SmallVectorImpl &iv) const; + bool processSizes(StatementContext &stmtCtx, + mlir::omp::SizesClauseOps &result) const; bool processDevice(lower::StatementContext &stmtCtx, mlir::omp::DeviceClauseOps &result) const; bool processDeviceType(mlir::omp::DeviceTypeClauseOps &result) const; diff --git a/flang/lib/Lower/OpenMP/OpenMP.cpp b/flang/lib/Lower/OpenMP/OpenMP.cpp index 1cb3335abbd06..9e56c2bfb7e25 100644 --- a/flang/lib/Lower/OpenMP/OpenMP.cpp +++ b/flang/lib/Lower/OpenMP/OpenMP.cpp @@ -1982,125 +1982,241 @@ genLoopOp(lower::AbstractConverter &converter, lower::SymMap &symTable, return loopOp; } -static mlir::omp::CanonicalLoopOp -genCanonicalLoopOp(lower::AbstractConverter &converter, lower::SymMap &symTable, - semantics::SemanticsContext &semaCtx, - lower::pft::Evaluation &eval, mlir::Location loc, - const ConstructQueue &queue, - ConstructQueue::const_iterator item, - llvm::ArrayRef ivs, - llvm::omp::Directive directive) { +static void genCanonicalLoopNest( + lower::AbstractConverter &converter, lower::SymMap &symTable, + semantics::SemanticsContext &semaCtx, lower::pft::Evaluation &eval, + mlir::Location loc, const ConstructQueue &queue, + ConstructQueue::const_iterator item, size_t numLoops, + llvm::SmallVectorImpl &loops) { + assert(loops.empty() && "Expecting empty list to fill"); + assert(numLoops >= 1 && "Expecting at least one loop"); + fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder(); - assert(ivs.size() == 1 && "Nested loops not yet implemented"); - const semantics::Symbol *iv = ivs[0]; + mlir::omp::LoopRelatedClauseOps loopInfo; + llvm::SmallVector ivs; + collectLoopRelatedInfo(converter, loc, eval, numLoops, loopInfo, ivs); + assert(ivs.size() == numLoops && + "Expected to parse as many loop variables as there are loops"); + + // Steps that follow: + // 1. Emit all of the loop's prologues (compute the tripcount) + // 2. Emit omp.canonical_loop nested inside each other (iteratively) + // 2.1. In the innermost omp.canonical_loop, emit the loop body prologue (in + // the body callback) + // + // Since emitting prologues and body code is split, remember prologue values + // for use when emitting the same loop's epilogues. + llvm::SmallVector tripcounts; + llvm::SmallVector clis; + llvm::SmallVector evals; + llvm::SmallVector loopVarTypes; + llvm::SmallVector loopStepVars; + llvm::SmallVector loopLBVars; + llvm::SmallVector blockArgs; + + // Step 1: Loop prologues + // Computing the trip count must happen before entering the outermost loop + lower::pft::Evaluation *innermostEval = &eval.getFirstNestedEvaluation(); + for ([[maybe_unused]] auto iv : ivs) { + if (innermostEval->getIf()->IsDoConcurrent()) { + // OpenMP specifies DO CONCURRENT only with the `!omp loop` construct. + // Will need to add special cases for this combination. + TODO(loc, "DO CONCURRENT as canonical loop not supported"); + } + + auto &doLoopEval = innermostEval->getFirstNestedEvaluation(); + evals.push_back(innermostEval); + + // Get the loop bounds (and increment) + // auto &doLoopEval = nestedEval.getFirstNestedEvaluation(); + auto *doStmt = doLoopEval.getIf(); + assert(doStmt && "Expected do loop to be in the nested evaluation"); + auto &loopControl = std::get>(doStmt->t); + assert(loopControl.has_value()); + auto *bounds = std::get_if(&loopControl->u); + assert(bounds && "Expected bounds for canonical loop"); + lower::StatementContext stmtCtx; + mlir::Value loopLBVar = fir::getBase( + converter.genExprValue(*semantics::GetExpr(bounds->lower), stmtCtx)); + mlir::Value loopUBVar = fir::getBase( + converter.genExprValue(*semantics::GetExpr(bounds->upper), stmtCtx)); + mlir::Value loopStepVar = [&]() { + if (bounds->step) { + return fir::getBase( + converter.genExprValue(*semantics::GetExpr(bounds->step), stmtCtx)); + } - auto &nestedEval = eval.getFirstNestedEvaluation(); - if (nestedEval.getIf()->IsDoConcurrent()) { - // OpenMP specifies DO CONCURRENT only with the `!omp loop` construct. Will - // need to add special cases for this combination. - TODO(loc, "DO CONCURRENT as canonical loop not supported"); + // If `step` is not present, assume it is `1`. + auto intTy = firOpBuilder.getI32Type(); + return firOpBuilder.createIntegerConstant(loc, intTy, 1); + }(); + + // Get the integer kind for the loop variable and cast the loop bounds + size_t loopVarTypeSize = bounds->name.thing.symbol->GetUltimate().size(); + mlir::Type loopVarType = getLoopVarType(converter, loopVarTypeSize); + loopVarTypes.push_back(loopVarType); + loopLBVar = firOpBuilder.createConvert(loc, loopVarType, loopLBVar); + loopUBVar = firOpBuilder.createConvert(loc, loopVarType, loopUBVar); + loopStepVar = firOpBuilder.createConvert(loc, loopVarType, loopStepVar); + loopLBVars.push_back(loopLBVar); + loopStepVars.push_back(loopStepVar); + + // Start lowering + mlir::Value zero = firOpBuilder.createIntegerConstant(loc, loopVarType, 0); + mlir::Value one = firOpBuilder.createIntegerConstant(loc, loopVarType, 1); + mlir::Value isDownwards = firOpBuilder.create( + loc, mlir::arith::CmpIPredicate::slt, loopStepVar, zero); + + // Ensure we are counting upwards. If not, negate step and swap lb and ub. + mlir::Value negStep = + firOpBuilder.create(loc, zero, loopStepVar); + mlir::Value incr = firOpBuilder.create( + loc, isDownwards, negStep, loopStepVar); + mlir::Value lb = firOpBuilder.create( + loc, isDownwards, loopUBVar, loopLBVar); + mlir::Value ub = firOpBuilder.create( + loc, isDownwards, loopLBVar, loopUBVar); + + // Compute the trip count assuming lb <= ub. This guarantees that the result + // is non-negative and we can use unsigned arithmetic. + mlir::Value span = firOpBuilder.create( + loc, ub, lb, ::mlir::arith::IntegerOverflowFlags::nuw); + mlir::Value tcMinusOne = + firOpBuilder.create(loc, span, incr); + mlir::Value tcIfLooping = firOpBuilder.create( + loc, tcMinusOne, one, ::mlir::arith::IntegerOverflowFlags::nuw); + + // Fall back to 0 if lb > ub + mlir::Value isZeroTC = firOpBuilder.create( + loc, mlir::arith::CmpIPredicate::slt, ub, lb); + mlir::Value tripcount = firOpBuilder.create( + loc, isZeroTC, zero, tcIfLooping); + tripcounts.push_back(tripcount); + + // Create the CLI handle. + auto newcli = firOpBuilder.create(loc); + mlir::Value cli = newcli.getResult(); + clis.push_back(cli); + + innermostEval = &*std::next(innermostEval->getNestedEvaluations().begin()); } - // Get the loop bounds (and increment) - auto &doLoopEval = nestedEval.getFirstNestedEvaluation(); - auto *doStmt = doLoopEval.getIf(); - assert(doStmt && "Expected do loop to be in the nested evaluation"); - auto &loopControl = std::get>(doStmt->t); - assert(loopControl.has_value()); - auto *bounds = std::get_if(&loopControl->u); - assert(bounds && "Expected bounds for canonical loop"); - lower::StatementContext stmtCtx; - mlir::Value loopLBVar = fir::getBase( - converter.genExprValue(*semantics::GetExpr(bounds->lower), stmtCtx)); - mlir::Value loopUBVar = fir::getBase( - converter.genExprValue(*semantics::GetExpr(bounds->upper), stmtCtx)); - mlir::Value loopStepVar = [&]() { - if (bounds->step) { - return fir::getBase( - converter.genExprValue(*semantics::GetExpr(bounds->step), stmtCtx)); - } + // Step 2: Create nested canoncial loops + for (auto i : llvm::seq(numLoops)) { + bool isInnermost = (i == numLoops - 1); + mlir::Type loopVarType = loopVarTypes[i]; + mlir::Value tripcount = tripcounts[i]; + mlir::Value cli = clis[i]; + auto &&eval = evals[i]; + + auto ivCallback = [&, i, isInnermost](mlir::Operation *op) + -> llvm::SmallVector { + mlir::Region ®ion = op->getRegion(0); + + // Create the op's region skeleton (BB taking the iv as argument) + firOpBuilder.createBlock(®ion, {}, {loopVarType}, {loc}); + blockArgs.push_back(region.front().getArgument(0)); + + // Step 2.1: Emit body prologue code + // Compute the translation from logical iteration number to the value of + // the loop's iteration variable only in the innermost body. Currently, + // loop transformations do not allow any instruction between loops, but + // this will change with + if (isInnermost) { + assert(blockArgs.size() == numLoops && + "Expecting all block args to have been collected by now"); + for (auto j : llvm::seq(numLoops)) { + mlir::Value natIterNum = fir::getBase(blockArgs[j]); + mlir::Value scaled = firOpBuilder.create( + loc, natIterNum, loopStepVars[j]); + mlir::Value userVal = firOpBuilder.create( + loc, loopLBVars[j], scaled); + + mlir::OpBuilder::InsertPoint insPt = + firOpBuilder.saveInsertionPoint(); + firOpBuilder.setInsertionPointToStart(firOpBuilder.getAllocaBlock()); + mlir::Type tempTy = converter.genType(*ivs[j]); + firOpBuilder.restoreInsertionPoint(insPt); + + // Write the loop value into loop variable + mlir::Value cvtVal = firOpBuilder.createConvert(loc, tempTy, userVal); + hlfir::Entity lhs{converter.getSymbolAddress(*ivs[j])}; + lhs = hlfir::derefPointersAndAllocatables(loc, firOpBuilder, lhs); + mlir::Operation *storeOp = + hlfir::AssignOp::create(firOpBuilder, loc, cvtVal, lhs); + firOpBuilder.setInsertionPointAfter(storeOp); + } + } - // If `step` is not present, assume it is `1`. - return firOpBuilder.createIntegerConstant(loc, firOpBuilder.getI32Type(), - 1); - }(); + return {ivs[i]}; + }; - // Get the integer kind for the loop variable and cast the loop bounds - size_t loopVarTypeSize = bounds->name.thing.symbol->GetUltimate().size(); - mlir::Type loopVarType = getLoopVarType(converter, loopVarTypeSize); - loopLBVar = firOpBuilder.createConvert(loc, loopVarType, loopLBVar); - loopUBVar = firOpBuilder.createConvert(loc, loopVarType, loopUBVar); - loopStepVar = firOpBuilder.createConvert(loc, loopVarType, loopStepVar); - - // Start lowering - mlir::Value zero = firOpBuilder.createIntegerConstant(loc, loopVarType, 0); - mlir::Value one = firOpBuilder.createIntegerConstant(loc, loopVarType, 1); - mlir::Value isDownwards = mlir::arith::CmpIOp::create( - firOpBuilder, loc, mlir::arith::CmpIPredicate::slt, loopStepVar, zero); - - // Ensure we are counting upwards. If not, negate step and swap lb and ub. - mlir::Value negStep = - mlir::arith::SubIOp::create(firOpBuilder, loc, zero, loopStepVar); - mlir::Value incr = mlir::arith::SelectOp::create( - firOpBuilder, loc, isDownwards, negStep, loopStepVar); - mlir::Value lb = mlir::arith::SelectOp::create(firOpBuilder, loc, isDownwards, - loopUBVar, loopLBVar); - mlir::Value ub = mlir::arith::SelectOp::create(firOpBuilder, loc, isDownwards, - loopLBVar, loopUBVar); - - // Compute the trip count assuming lb <= ub. This guarantees that the result - // is non-negative and we can use unsigned arithmetic. - mlir::Value span = mlir::arith::SubIOp::create( - firOpBuilder, loc, ub, lb, ::mlir::arith::IntegerOverflowFlags::nuw); - mlir::Value tcMinusOne = - mlir::arith::DivUIOp::create(firOpBuilder, loc, span, incr); - mlir::Value tcIfLooping = - mlir::arith::AddIOp::create(firOpBuilder, loc, tcMinusOne, one, - ::mlir::arith::IntegerOverflowFlags::nuw); - - // Fall back to 0 if lb > ub - mlir::Value isZeroTC = mlir::arith::CmpIOp::create( - firOpBuilder, loc, mlir::arith::CmpIPredicate::slt, ub, lb); - mlir::Value tripcount = mlir::arith::SelectOp::create( - firOpBuilder, loc, isZeroTC, zero, tcIfLooping); - - // Create the CLI handle. - auto newcli = mlir::omp::NewCliOp::create(firOpBuilder, loc); - mlir::Value cli = newcli.getResult(); - - auto ivCallback = [&](mlir::Operation *op) - -> llvm::SmallVector { - mlir::Region ®ion = op->getRegion(0); - - // Create the op's region skeleton (BB taking the iv as argument) - firOpBuilder.createBlock(®ion, {}, {loopVarType}, {loc}); - - // Compute the value of the loop variable from the logical iteration number. - mlir::Value natIterNum = fir::getBase(region.front().getArgument(0)); - mlir::Value scaled = - mlir::arith::MulIOp::create(firOpBuilder, loc, natIterNum, loopStepVar); - mlir::Value userVal = - mlir::arith::AddIOp::create(firOpBuilder, loc, loopLBVar, scaled); - - // Write loop value to loop variable - mlir::Operation *storeOp = setLoopVar(converter, loc, userVal, iv); - - firOpBuilder.setInsertionPointAfter(storeOp); - return {iv}; - }; + // Create the omp.canonical_loop operation + auto opGenInfo = OpWithBodyGenInfo(converter, symTable, semaCtx, loc, *eval, + llvm::omp::Directive::OMPD_unknown) + .setGenSkeletonOnly(!isInnermost) + .setClauses(&item->clauses) + .setPrivatize(false) + .setGenRegionEntryCb(ivCallback); + auto canonLoop = genOpWithBody( + std::move(opGenInfo), queue, item, tripcount, cli); + loops.push_back(canonLoop); + + // Insert next loop nested inside last loop + firOpBuilder.setInsertionPoint( + canonLoop.getRegion().back().getTerminator()); + } - // Create the omp.canonical_loop operation - auto canonLoop = genOpWithBody( - OpWithBodyGenInfo(converter, symTable, semaCtx, loc, nestedEval, - directive) - .setClauses(&item->clauses) - .setPrivatize(false) - .setGenRegionEntryCb(ivCallback), - queue, item, tripcount, cli); + firOpBuilder.setInsertionPointAfter(loops.front()); +} + +static void genTileOp(Fortran::lower::AbstractConverter &converter, + Fortran::lower::SymMap &symTable, + lower::StatementContext &stmtCtx, + Fortran::semantics::SemanticsContext &semaCtx, + Fortran::lower::pft::Evaluation &eval, mlir::Location loc, + const ConstructQueue &queue, + ConstructQueue::const_iterator item) { + fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder(); - firOpBuilder.setInsertionPointAfter(canonLoop); - return canonLoop; + mlir::omp::SizesClauseOps sizesClause; + ClauseProcessor cp(converter, semaCtx, item->clauses); + cp.processSizes(stmtCtx, sizesClause); + + size_t numLoops = sizesClause.sizes.size(); + llvm::SmallVector canonLoops; + canonLoops.reserve(numLoops); + + genCanonicalLoopNest(converter, symTable, semaCtx, eval, loc, queue, item, + numLoops, canonLoops); + assert((canonLoops.size() == numLoops) && + "Expecting the predetermined number of loops"); + + llvm::SmallVector applyees; + applyees.reserve(numLoops); + for (mlir::omp::CanonicalLoopOp l : canonLoops) + applyees.push_back(l.getCli()); + + // Emit the associated loops and create a CLI for each affected loop + llvm::SmallVector gridGeneratees; + llvm::SmallVector intratileGeneratees; + gridGeneratees.reserve(numLoops); + intratileGeneratees.reserve(numLoops); + for ([[maybe_unused]] auto i : llvm::seq(0, sizesClause.sizes.size())) { + auto gridCLI = firOpBuilder.create(loc); + gridGeneratees.push_back(gridCLI.getResult()); + auto intratileCLI = firOpBuilder.create(loc); + intratileGeneratees.push_back(intratileCLI.getResult()); + } + + llvm::SmallVector generatees; + generatees.reserve(2 * numLoops); + generatees.append(gridGeneratees); + generatees.append(intratileGeneratees); + + firOpBuilder.create(loc, generatees, applyees, + sizesClause.sizes); } static void genUnrollOp(Fortran::lower::AbstractConverter &converter, @@ -2112,22 +2228,22 @@ static void genUnrollOp(Fortran::lower::AbstractConverter &converter, ConstructQueue::const_iterator item) { fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder(); - mlir::omp::LoopRelatedClauseOps loopInfo; - llvm::SmallVector iv; - collectLoopRelatedInfo(converter, loc, eval, item->clauses, loopInfo, iv); - // Clauses for unrolling not yet implemnted ClauseProcessor cp(converter, semaCtx, item->clauses); cp.processTODO( loc, llvm::omp::Directive::OMPD_unroll); // Emit the associated loop - auto canonLoop = - genCanonicalLoopOp(converter, symTable, semaCtx, eval, loc, queue, item, - iv, llvm::omp::Directive::OMPD_unroll); + llvm::SmallVector canonLoops; + genCanonicalLoopNest(converter, symTable, semaCtx, eval, loc, queue, item, 1, + canonLoops); + + llvm::SmallVector applyees; + for (auto &&canonLoop : canonLoops) + applyees.push_back(canonLoop.getCli()); // Apply unrolling to it - auto cli = canonLoop.getCli(); + auto cli = llvm::getSingleElement(canonLoops).getCli(); mlir::omp::UnrollHeuristicOp::create(firOpBuilder, loc, cli); } @@ -3360,13 +3476,9 @@ static void genOMPDispatch(lower::AbstractConverter &converter, newOp = genTeamsOp(converter, symTable, stmtCtx, semaCtx, eval, loc, queue, item); break; - case llvm::omp::Directive::OMPD_tile: { - unsigned version = semaCtx.langOptions().OpenMPVersion; - if (!semaCtx.langOptions().OpenMPSimd) - TODO(loc, "Unhandled loop directive (" + - llvm::omp::getOpenMPDirectiveName(dir, version) + ")"); + case llvm::omp::Directive::OMPD_tile: + genTileOp(converter, symTable, stmtCtx, semaCtx, eval, loc, queue, item); break; - } case llvm::omp::Directive::OMPD_unroll: genUnrollOp(converter, symTable, stmtCtx, semaCtx, eval, loc, queue, item); break; diff --git a/flang/lib/Lower/OpenMP/Utils.cpp b/flang/lib/Lower/OpenMP/Utils.cpp index 83b7ccb1ce0ee..29cccbd1bfe5a 100644 --- a/flang/lib/Lower/OpenMP/Utils.cpp +++ b/flang/lib/Lower/OpenMP/Utils.cpp @@ -652,7 +652,6 @@ int64_t collectLoopRelatedInfo( mlir::omp::LoopRelatedClauseOps &result, llvm::SmallVectorImpl &iv) { int64_t numCollapse = 1; - fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder(); // Collect the loops to collapse. lower::pft::Evaluation *doConstructEval = &eval.getFirstNestedEvaluation(); @@ -667,6 +666,25 @@ int64_t collectLoopRelatedInfo( numCollapse = collapseValue; } + collectLoopRelatedInfo(converter, currentLocation, eval, numCollapse, result, + iv); + return numCollapse; +} + +void collectLoopRelatedInfo( + lower::AbstractConverter &converter, mlir::Location currentLocation, + lower::pft::Evaluation &eval, int64_t numCollapse, + mlir::omp::LoopRelatedClauseOps &result, + llvm::SmallVectorImpl &iv) { + + fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder(); + + // Collect the loops to collapse. + lower::pft::Evaluation *doConstructEval = &eval.getFirstNestedEvaluation(); + if (doConstructEval->getIf()->IsDoConcurrent()) { + TODO(currentLocation, "Do Concurrent in Worksharing loop construct"); + } + // Collect sizes from tile directive if present. std::int64_t sizesLengthValue = 0l; if (auto *ompCons{eval.getIf()}) { @@ -676,7 +694,7 @@ int64_t collectLoopRelatedInfo( }); } - collapseValue = std::max(collapseValue, sizesLengthValue); + std::int64_t collapseValue = std::max(numCollapse, sizesLengthValue); std::size_t loopVarTypeSize = 0; do { lower::pft::Evaluation *doLoop = @@ -709,8 +727,6 @@ int64_t collectLoopRelatedInfo( } while (collapseValue > 0); convertLoopBounds(converter, currentLocation, result, loopVarTypeSize); - - return numCollapse; } } // namespace omp diff --git a/flang/lib/Lower/OpenMP/Utils.h b/flang/lib/Lower/OpenMP/Utils.h index 5f191d89ae205..69499f9c7b621 100644 --- a/flang/lib/Lower/OpenMP/Utils.h +++ b/flang/lib/Lower/OpenMP/Utils.h @@ -165,6 +165,13 @@ int64_t collectLoopRelatedInfo( mlir::omp::LoopRelatedClauseOps &result, llvm::SmallVectorImpl &iv); +void collectLoopRelatedInfo( + lower::AbstractConverter &converter, mlir::Location currentLocation, + lower::pft::Evaluation &eval, std::int64_t collapseValue, + // const omp::List &clauses, + mlir::omp::LoopRelatedClauseOps &result, + llvm::SmallVectorImpl &iv); + void collectTileSizesFromOpenMPConstruct( const parser::OpenMPConstruct *ompCons, llvm::SmallVectorImpl &tileSizes, diff --git a/flang/lib/Semantics/check-directive-structure.h b/flang/lib/Semantics/check-directive-structure.h index b1bf3e550aebc..bd78d3cfe91e7 100644 --- a/flang/lib/Semantics/check-directive-structure.h +++ b/flang/lib/Semantics/check-directive-structure.h @@ -383,7 +383,8 @@ class DirectiveStructureChecker : public virtual BaseChecker { const C &clause, const parser::ScalarIntConstantExpr &i); void RequiresPositiveParameter(const C &clause, - const parser::ScalarIntExpr &i, llvm::StringRef paramName = "parameter"); + const parser::ScalarIntExpr &i, llvm::StringRef paramName = "parameter", + bool allowZero = true); void OptionalConstantPositiveParameter( const C &clause, const std::optional &o); @@ -657,9 +658,9 @@ void DirectiveStructureChecker::SayNotMatching( template void DirectiveStructureChecker::RequiresPositiveParameter(const C &clause, - const parser::ScalarIntExpr &i, llvm::StringRef paramName) { + const parser::ScalarIntExpr &i, llvm::StringRef paramName, bool allowZero) { if (const auto v{GetIntValue(i)}) { - if (*v < 0) { + if (*v < (allowZero ? 0 : 1)) { context_.Say(GetContext().clauseSource, "The %s of the %s clause must be " "a positive integer expression"_err_en_US, diff --git a/flang/lib/Semantics/check-omp-structure.cpp b/flang/lib/Semantics/check-omp-structure.cpp index 1f059f747bad0..c0c41c1eed89d 100644 --- a/flang/lib/Semantics/check-omp-structure.cpp +++ b/flang/lib/Semantics/check-omp-structure.cpp @@ -3145,6 +3145,13 @@ void OmpStructureChecker::Enter(const parser::OmpClause &x) { } } +void OmpStructureChecker::Enter(const parser::OmpClause::Sizes &c) { + CheckAllowedClause(llvm::omp::Clause::OMPC_sizes); + for (const parser::Cosubscript &v : c.v) + RequiresPositiveParameter(llvm::omp::Clause::OMPC_sizes, v, + /*paramName=*/"parameter", /*allowZero=*/false); +} + // Following clauses do not have a separate node in parse-tree.h. CHECK_SIMPLE_CLAUSE(Absent, OMPC_absent) CHECK_SIMPLE_CLAUSE(Affinity, OMPC_affinity) @@ -3186,7 +3193,6 @@ CHECK_SIMPLE_CLAUSE(Notinbranch, OMPC_notinbranch) CHECK_SIMPLE_CLAUSE(Partial, OMPC_partial) CHECK_SIMPLE_CLAUSE(ProcBind, OMPC_proc_bind) CHECK_SIMPLE_CLAUSE(Simd, OMPC_simd) -CHECK_SIMPLE_CLAUSE(Sizes, OMPC_sizes) CHECK_SIMPLE_CLAUSE(Permutation, OMPC_permutation) CHECK_SIMPLE_CLAUSE(Uniform, OMPC_uniform) CHECK_SIMPLE_CLAUSE(Unknown, OMPC_unknown) diff --git a/flang/lib/Semantics/resolve-directives.cpp b/flang/lib/Semantics/resolve-directives.cpp index 624b89005c809..02fcf021137ab 100644 --- a/flang/lib/Semantics/resolve-directives.cpp +++ b/flang/lib/Semantics/resolve-directives.cpp @@ -2421,10 +2421,18 @@ void OmpAttributeVisitor::PrivatizeAssociatedLoopIndexAndCheckLoopLevel( void OmpAttributeVisitor::CheckAssocLoopLevel( std::int64_t level, const parser::OmpClause *clause) { if (clause && level != 0) { - context_.Say(clause->source, - "The value of the parameter in the COLLAPSE or ORDERED clause must" - " not be larger than the number of nested loops" - " following the construct."_err_en_US); + switch (clause->Id()) { + case llvm::omp::OMPC_sizes: + context_.Say(clause->source, + "The SIZES clause has more entries than there are nested canonical loops."_err_en_US); + break; + default: + context_.Say(clause->source, + "The value of the parameter in the COLLAPSE or ORDERED clause must" + " not be larger than the number of nested loops" + " following the construct."_err_en_US); + break; + } } } diff --git a/flang/test/Lower/OpenMP/tile01.f90 b/flang/test/Lower/OpenMP/tile01.f90 new file mode 100644 index 0000000000000..7603eee4b18d8 --- /dev/null +++ b/flang/test/Lower/OpenMP/tile01.f90 @@ -0,0 +1,58 @@ +! RUN: %flang_fc1 -emit-hlfir -fopenmp -fopenmp-version=51 -o - %s | FileCheck %s + + +subroutine omp_tile01(lb, ub, inc) + integer res, i, lb, ub, inc + + !$omp tile sizes(4) + do i = lb, ub, inc + res = i + end do + !$omp end tile + +end subroutine omp_tile01 + + +! CHECK: func.func @_QPomp_tile01( +! CHECK: %[[ARG0:.*]]: !fir.ref {fir.bindc_name = "lb"}, +! CHECK: %[[ARG1:.*]]: !fir.ref {fir.bindc_name = "ub"}, +! CHECK: %[[ARG2:.*]]: !fir.ref {fir.bindc_name = "inc"}) { +! CHECK: %[[VAL_0:.*]] = fir.dummy_scope : !fir.dscope +! CHECK: %[[VAL_1:.*]] = fir.alloca i32 {bindc_name = "i", uniq_name = "_QFomp_tile01Ei"} +! CHECK: %[[VAL_2:.*]]:2 = hlfir.declare %[[VAL_1]] {uniq_name = "_QFomp_tile01Ei"} : (!fir.ref) -> (!fir.ref, !fir.ref) +! CHECK: %[[VAL_3:.*]]:2 = hlfir.declare %[[ARG2]] dummy_scope %[[VAL_0]] {uniq_name = "_QFomp_tile01Einc"} : (!fir.ref, !fir.dscope) -> (!fir.ref, !fir.ref) +! CHECK: %[[VAL_4:.*]]:2 = hlfir.declare %[[ARG0]] dummy_scope %[[VAL_0]] {uniq_name = "_QFomp_tile01Elb"} : (!fir.ref, !fir.dscope) -> (!fir.ref, !fir.ref) +! CHECK: %[[VAL_5:.*]] = fir.alloca i32 {bindc_name = "res", uniq_name = "_QFomp_tile01Eres"} +! CHECK: %[[VAL_6:.*]]:2 = hlfir.declare %[[VAL_5]] {uniq_name = "_QFomp_tile01Eres"} : (!fir.ref) -> (!fir.ref, !fir.ref) +! CHECK: %[[VAL_7:.*]]:2 = hlfir.declare %[[ARG1]] dummy_scope %[[VAL_0]] {uniq_name = "_QFomp_tile01Eub"} : (!fir.ref, !fir.dscope) -> (!fir.ref, !fir.ref) +! CHECK: %[[VAL_8:.*]] = arith.constant 4 : i32 +! CHECK: %[[VAL_9:.*]] = fir.load %[[VAL_4]]#0 : !fir.ref +! CHECK: %[[VAL_10:.*]] = fir.load %[[VAL_7]]#0 : !fir.ref +! CHECK: %[[VAL_11:.*]] = fir.load %[[VAL_3]]#0 : !fir.ref +! CHECK: %[[VAL_12:.*]] = arith.constant 0 : i32 +! CHECK: %[[VAL_13:.*]] = arith.constant 1 : i32 +! CHECK: %[[VAL_14:.*]] = arith.cmpi slt, %[[VAL_11]], %[[VAL_12]] : i32 +! CHECK: %[[VAL_15:.*]] = arith.subi %[[VAL_12]], %[[VAL_11]] : i32 +! CHECK: %[[VAL_16:.*]] = arith.select %[[VAL_14]], %[[VAL_15]], %[[VAL_11]] : i32 +! CHECK: %[[VAL_17:.*]] = arith.select %[[VAL_14]], %[[VAL_10]], %[[VAL_9]] : i32 +! CHECK: %[[VAL_18:.*]] = arith.select %[[VAL_14]], %[[VAL_9]], %[[VAL_10]] : i32 +! CHECK: %[[VAL_19:.*]] = arith.subi %[[VAL_18]], %[[VAL_17]] overflow : i32 +! CHECK: %[[VAL_20:.*]] = arith.divui %[[VAL_19]], %[[VAL_16]] : i32 +! CHECK: %[[VAL_21:.*]] = arith.addi %[[VAL_20]], %[[VAL_13]] overflow : i32 +! CHECK: %[[VAL_22:.*]] = arith.cmpi slt, %[[VAL_18]], %[[VAL_17]] : i32 +! CHECK: %[[VAL_23:.*]] = arith.select %[[VAL_22]], %[[VAL_12]], %[[VAL_21]] : i32 +! CHECK: %[[VAL_24:.*]] = omp.new_cli +! CHECK: omp.canonical_loop(%[[VAL_24]]) %[[VAL_25:.*]] : i32 in range(%[[VAL_23]]) { +! CHECK: %[[VAL_26:.*]] = arith.muli %[[VAL_25]], %[[VAL_11]] : i32 +! CHECK: %[[VAL_27:.*]] = arith.addi %[[VAL_9]], %[[VAL_26]] : i32 +! CHECK: hlfir.assign %[[VAL_27]] to %[[VAL_2]]#0 : i32, !fir.ref +! CHECK: %[[VAL_28:.*]] = fir.load %[[VAL_2]]#0 : !fir.ref +! CHECK: hlfir.assign %[[VAL_28]] to %[[VAL_6]]#0 : i32, !fir.ref +! CHECK: omp.terminator +! CHECK: } +! CHECK: %[[VAL_29:.*]] = omp.new_cli +! CHECK: %[[VAL_30:.*]] = omp.new_cli +! CHECK: omp.tile (%[[VAL_29]], %[[VAL_30]]) <- (%[[VAL_24]]) sizes(%[[VAL_8]] : i32) +! CHECK: return +! CHECK: } + diff --git a/flang/test/Lower/OpenMP/tile02.f90 b/flang/test/Lower/OpenMP/tile02.f90 new file mode 100644 index 0000000000000..5df506d17ed05 --- /dev/null +++ b/flang/test/Lower/OpenMP/tile02.f90 @@ -0,0 +1,88 @@ +! RUN: %flang_fc1 -emit-hlfir -fopenmp -fopenmp-version=51 -o - %s | FileCheck %s + + +subroutine omp_tile02(lb, ub, inc) + integer res, i, lb, ub, inc + + !$omp tile sizes(3,7) + do i = lb, ub, inc + do j = lb, ub, inc + res = i + j + end do + end do + !$omp end tile + +end subroutine omp_tile02 + + +! CHECK: func.func @_QPomp_tile02( +! CHECK: %[[ARG0:.*]]: !fir.ref {fir.bindc_name = "lb"}, +! CHECK: %[[ARG1:.*]]: !fir.ref {fir.bindc_name = "ub"}, +! CHECK: %[[ARG2:.*]]: !fir.ref {fir.bindc_name = "inc"}) { +! CHECK: %[[VAL_0:.*]] = fir.dummy_scope : !fir.dscope +! CHECK: %[[VAL_1:.*]] = fir.alloca i32 {bindc_name = "i", uniq_name = "_QFomp_tile02Ei"} +! CHECK: %[[VAL_2:.*]]:2 = hlfir.declare %[[VAL_1]] {uniq_name = "_QFomp_tile02Ei"} : (!fir.ref) -> (!fir.ref, !fir.ref) +! CHECK: %[[VAL_3:.*]]:2 = hlfir.declare %[[ARG2]] dummy_scope %[[VAL_0]] {uniq_name = "_QFomp_tile02Einc"} : (!fir.ref, !fir.dscope) -> (!fir.ref, !fir.ref) +! CHECK: %[[VAL_4:.*]] = fir.alloca i32 {bindc_name = "j", uniq_name = "_QFomp_tile02Ej"} +! CHECK: %[[VAL_5:.*]]:2 = hlfir.declare %[[VAL_4]] {uniq_name = "_QFomp_tile02Ej"} : (!fir.ref) -> (!fir.ref, !fir.ref) +! CHECK: %[[VAL_6:.*]]:2 = hlfir.declare %[[ARG0]] dummy_scope %[[VAL_0]] {uniq_name = "_QFomp_tile02Elb"} : (!fir.ref, !fir.dscope) -> (!fir.ref, !fir.ref) +! CHECK: %[[VAL_7:.*]] = fir.alloca i32 {bindc_name = "res", uniq_name = "_QFomp_tile02Eres"} +! CHECK: %[[VAL_8:.*]]:2 = hlfir.declare %[[VAL_7]] {uniq_name = "_QFomp_tile02Eres"} : (!fir.ref) -> (!fir.ref, !fir.ref) +! CHECK: %[[VAL_9:.*]]:2 = hlfir.declare %[[ARG1]] dummy_scope %[[VAL_0]] {uniq_name = "_QFomp_tile02Eub"} : (!fir.ref, !fir.dscope) -> (!fir.ref, !fir.ref) +! CHECK: %[[VAL_10:.*]] = arith.constant 3 : i32 +! CHECK: %[[VAL_11:.*]] = arith.constant 7 : i32 +! CHECK: %[[VAL_12:.*]] = fir.load %[[VAL_6]]#0 : !fir.ref +! CHECK: %[[VAL_13:.*]] = fir.load %[[VAL_9]]#0 : !fir.ref +! CHECK: %[[VAL_14:.*]] = fir.load %[[VAL_3]]#0 : !fir.ref +! CHECK: %[[VAL_15:.*]] = arith.constant 0 : i32 +! CHECK: %[[VAL_16:.*]] = arith.constant 1 : i32 +! CHECK: %[[VAL_17:.*]] = arith.cmpi slt, %[[VAL_14]], %[[VAL_15]] : i32 +! CHECK: %[[VAL_18:.*]] = arith.subi %[[VAL_15]], %[[VAL_14]] : i32 +! CHECK: %[[VAL_19:.*]] = arith.select %[[VAL_17]], %[[VAL_18]], %[[VAL_14]] : i32 +! CHECK: %[[VAL_20:.*]] = arith.select %[[VAL_17]], %[[VAL_13]], %[[VAL_12]] : i32 +! CHECK: %[[VAL_21:.*]] = arith.select %[[VAL_17]], %[[VAL_12]], %[[VAL_13]] : i32 +! CHECK: %[[VAL_22:.*]] = arith.subi %[[VAL_21]], %[[VAL_20]] overflow : i32 +! CHECK: %[[VAL_23:.*]] = arith.divui %[[VAL_22]], %[[VAL_19]] : i32 +! CHECK: %[[VAL_24:.*]] = arith.addi %[[VAL_23]], %[[VAL_16]] overflow : i32 +! CHECK: %[[VAL_25:.*]] = arith.cmpi slt, %[[VAL_21]], %[[VAL_20]] : i32 +! CHECK: %[[VAL_26:.*]] = arith.select %[[VAL_25]], %[[VAL_15]], %[[VAL_24]] : i32 +! CHECK: %[[VAL_27:.*]] = omp.new_cli +! CHECK: %[[VAL_28:.*]] = fir.load %[[VAL_6]]#0 : !fir.ref +! CHECK: %[[VAL_29:.*]] = fir.load %[[VAL_9]]#0 : !fir.ref +! CHECK: %[[VAL_30:.*]] = fir.load %[[VAL_3]]#0 : !fir.ref +! CHECK: %[[VAL_31:.*]] = arith.constant 0 : i32 +! CHECK: %[[VAL_32:.*]] = arith.constant 1 : i32 +! CHECK: %[[VAL_33:.*]] = arith.cmpi slt, %[[VAL_30]], %[[VAL_31]] : i32 +! CHECK: %[[VAL_34:.*]] = arith.subi %[[VAL_31]], %[[VAL_30]] : i32 +! CHECK: %[[VAL_35:.*]] = arith.select %[[VAL_33]], %[[VAL_34]], %[[VAL_30]] : i32 +! CHECK: %[[VAL_36:.*]] = arith.select %[[VAL_33]], %[[VAL_29]], %[[VAL_28]] : i32 +! CHECK: %[[VAL_37:.*]] = arith.select %[[VAL_33]], %[[VAL_28]], %[[VAL_29]] : i32 +! CHECK: %[[VAL_38:.*]] = arith.subi %[[VAL_37]], %[[VAL_36]] overflow : i32 +! CHECK: %[[VAL_39:.*]] = arith.divui %[[VAL_38]], %[[VAL_35]] : i32 +! CHECK: %[[VAL_40:.*]] = arith.addi %[[VAL_39]], %[[VAL_32]] overflow : i32 +! CHECK: %[[VAL_41:.*]] = arith.cmpi slt, %[[VAL_37]], %[[VAL_36]] : i32 +! CHECK: %[[VAL_42:.*]] = arith.select %[[VAL_41]], %[[VAL_31]], %[[VAL_40]] : i32 +! CHECK: %[[VAL_43:.*]] = omp.new_cli +! CHECK: omp.canonical_loop(%[[VAL_27]]) %[[VAL_44:.*]] : i32 in range(%[[VAL_26]]) { +! CHECK: omp.canonical_loop(%[[VAL_43]]) %[[VAL_45:.*]] : i32 in range(%[[VAL_42]]) { +! CHECK: %[[VAL_46:.*]] = arith.muli %[[VAL_44]], %[[VAL_14]] : i32 +! CHECK: %[[VAL_47:.*]] = arith.addi %[[VAL_12]], %[[VAL_46]] : i32 +! CHECK: hlfir.assign %[[VAL_47]] to %[[VAL_2]]#0 : i32, !fir.ref +! CHECK: %[[VAL_48:.*]] = arith.muli %[[VAL_45]], %[[VAL_30]] : i32 +! CHECK: %[[VAL_49:.*]] = arith.addi %[[VAL_28]], %[[VAL_48]] : i32 +! CHECK: hlfir.assign %[[VAL_49]] to %[[VAL_5]]#0 : i32, !fir.ref +! CHECK: %[[VAL_50:.*]] = fir.load %[[VAL_2]]#0 : !fir.ref +! CHECK: %[[VAL_51:.*]] = fir.load %[[VAL_5]]#0 : !fir.ref +! CHECK: %[[VAL_52:.*]] = arith.addi %[[VAL_50]], %[[VAL_51]] : i32 +! CHECK: hlfir.assign %[[VAL_52]] to %[[VAL_8]]#0 : i32, !fir.ref +! CHECK: omp.terminator +! CHECK: } +! CHECK: omp.terminator +! CHECK: } +! CHECK: %[[VAL_53:.*]] = omp.new_cli +! CHECK: %[[VAL_54:.*]] = omp.new_cli +! CHECK: %[[VAL_55:.*]] = omp.new_cli +! CHECK: %[[VAL_56:.*]] = omp.new_cli +! CHECK: omp.tile (%[[VAL_53]], %[[VAL_55]], %[[VAL_54]], %[[VAL_56]]) <- (%[[VAL_27]], %[[VAL_43]]) sizes(%[[VAL_10]], %[[VAL_11]] : i32, i32) +! CHECK: return +! CHECK: } diff --git a/flang/test/Parser/OpenMP/loop-transformation-construct02.f90 b/flang/test/Parser/OpenMP/loop-transformation-construct02.f90 index a6af35a0111a3..a876c77a274b5 100644 --- a/flang/test/Parser/OpenMP/loop-transformation-construct02.f90 +++ b/flang/test/Parser/OpenMP/loop-transformation-construct02.f90 @@ -11,7 +11,7 @@ subroutine loop_transformation_construct !$omp do !$omp unroll - !$omp tile + !$omp tile sizes(2) do i = 1, I y(i) = y(i) * 5 end do @@ -34,7 +34,8 @@ subroutine loop_transformation_construct !CHECK-PARSE-NEXT: | | | | OpenMPLoopConstruct !CHECK-PARSE-NEXT: | | | | | OmpBeginLoopDirective !CHECK-PARSE-NEXT: | | | | | | OmpDirectiveName -> llvm::omp::Directive = tile -!CHECK-PARSE-NEXT: | | | | | | OmpClauseList -> +!CHECK-PARSE-NEXT: | | | | | | OmpClauseList -> OmpClause -> Sizes -> Scalar -> Integer -> Expr = '2_4' +!CHECK-PARSE-NEXT: | | | | | | | LiteralConstant -> IntLiteralConstant = '2' !CHECK-PARSE-NEXT: | | | | | | Flags = None !CHECK-PARSE-NEXT: | | | | | DoConstruct !CHECK-PARSE-NEXT: | | | | | | NonLabelDoStmt diff --git a/flang/test/Parser/OpenMP/tile-fail.f90 b/flang/test/Parser/OpenMP/tile-fail.f90 new file mode 100644 index 0000000000000..0a92e5bcb6570 --- /dev/null +++ b/flang/test/Parser/OpenMP/tile-fail.f90 @@ -0,0 +1,32 @@ +! RUN: split-file %s %t +! RUN: not %flang_fc1 -fsyntax-only -fopenmp %t/stray_end1.f90 2>&1 | FileCheck %t/stray_end1.f90 +! RUN: not %flang_fc1 -fsyntax-only -fopenmp %t/stray_end2.f90 2>&1 | FileCheck %t/stray_end2.f90 +! RUN: not %flang_fc1 -fsyntax-only -fopenmp %t/stray_begin.f90 2>&1 | FileCheck %t/stray_begin.f90 + + +!--- stray_end1.f90 +! Parser error + +subroutine stray_end1 + !CHECK: error: expected OpenMP construct + !$omp end tile +end subroutine + + +!--- stray_end2.f90 +! Semantic error + +subroutine stray_end2 + print * + !CHECK: error: The END TILE directive must follow the DO loop associated with the loop construct + !$omp end tile +end subroutine + + +!--- stray_begin.f90 + +subroutine stray_begin + !CHECK: error: A DO loop must follow the TILE directive + !$omp tile sizes(2) +end subroutine + diff --git a/flang/test/Parser/OpenMP/tile.f90 b/flang/test/Parser/OpenMP/tile.f90 index 2ea17471866a4..82004fd37a0f2 100644 --- a/flang/test/Parser/OpenMP/tile.f90 +++ b/flang/test/Parser/OpenMP/tile.f90 @@ -1,12 +1,12 @@ -! RUN: %flang_fc1 -fdebug-unparse -fopenmp %s | FileCheck --ignore-case %s -! RUN: %flang_fc1 -fdebug-dump-parse-tree -fopenmp %s | FileCheck --check-prefix="PARSE-TREE" %s +! RUN: %flang_fc1 -fdebug-unparse -fopenmp -fopenmp-version=51 %s | FileCheck --ignore-case %s +! RUN: %flang_fc1 -fdebug-dump-parse-tree -fopenmp -fopenmp-version=51 %s | FileCheck --check-prefix="PARSE-TREE" %s subroutine openmp_tiles(x) integer, intent(inout)::x -!CHECK: !$omp tile -!$omp tile +!CHECK: !$omp tile sizes(2_4) +!$omp tile sizes(2) !CHECK: do do x = 1, 100 call F1() @@ -17,7 +17,12 @@ subroutine openmp_tiles(x) !PARSE-TREE: OpenMPConstruct -> OpenMPLoopConstruct !PARSE-TREE: OmpBeginLoopDirective +!PARSE-TREE: OmpClauseList -> OmpClause -> Sizes -> Scalar -> Integer -> Expr = '2_4' +!PARSE-TREE: LiteralConstant -> IntLiteralConstant = '2' +!PARSE-TREE: Flags = None +!PARSE-TREE: DoConstruct +!PARSE-TREE: EndDoStmt +!PARSE-TREE: OmpEndLoopDirective !PARSE-TREE: OmpDirectiveName -> llvm::omp::Directive = tile END subroutine openmp_tiles - diff --git a/flang/test/Semantics/OpenMP/tile01.f90 b/flang/test/Semantics/OpenMP/tile01.f90 new file mode 100644 index 0000000000000..3d7b3f4f42e92 --- /dev/null +++ b/flang/test/Semantics/OpenMP/tile01.f90 @@ -0,0 +1,26 @@ +! Testing the Semantics of tile +!RUN: %python %S/../test_errors.py %s %flang -fopenmp -fopenmp-version=51 + + +subroutine missing_sizes + implicit none + integer i + + !ERROR: At least one of SIZES clause must appear on the TILE directive + !$omp tile + do i = 1, 42 + print *, i + end do +end subroutine + + +subroutine double_sizes + implicit none + integer i + + !ERROR: At most one SIZES clause can appear on the TILE directive + !$omp tile sizes(2) sizes(2) + do i = 1, 5 + print *, i + end do +end subroutine diff --git a/flang/test/Semantics/OpenMP/tile02.f90 b/flang/test/Semantics/OpenMP/tile02.f90 new file mode 100644 index 0000000000000..676796375353f --- /dev/null +++ b/flang/test/Semantics/OpenMP/tile02.f90 @@ -0,0 +1,15 @@ +! Testing the Semantics of tile +!RUN: %python %S/../test_errors.py %s %flang -fopenmp -fopenmp-version=51 + + +subroutine on_unroll + implicit none + integer i + + !ERROR: If a loop construct has been fully unrolled, it cannot then be tiled + !$omp tile sizes(2) + !$omp unroll + do i = 1, 5 + print *, i + end do +end subroutine diff --git a/flang/test/Semantics/OpenMP/tile03.f90 b/flang/test/Semantics/OpenMP/tile03.f90 new file mode 100644 index 0000000000000..e5c134638ac8d --- /dev/null +++ b/flang/test/Semantics/OpenMP/tile03.f90 @@ -0,0 +1,15 @@ +! Testing the Semantics of tile +!RUN: %python %S/../test_errors.py %s %flang -fopenmp -fopenmp-version=51 + + +subroutine loop_assoc + implicit none + integer :: i = 0 + + !$omp tile sizes(2) + !ERROR: The associated loop of a loop-associated directive cannot be a DO WHILE. + do while (i <= 10) + i = i + 1 + print *, i + end do +end subroutine diff --git a/flang/test/Semantics/OpenMP/tile04.f90 b/flang/test/Semantics/OpenMP/tile04.f90 new file mode 100644 index 0000000000000..2b503efbcf52b --- /dev/null +++ b/flang/test/Semantics/OpenMP/tile04.f90 @@ -0,0 +1,38 @@ +! Testing the Semantics of tile +!RUN: %python %S/../test_errors.py %s %flang -fopenmp -fopenmp-version=51 + + +subroutine threads_zero + implicit none + integer i + + !ERROR: The parameter of the NUM_THREADS clause must be a positive integer expression + !$omp parallel do num_threads(-1) + do i = 1, 5 + print *, i + end do +end subroutine + + +subroutine sizes_zero + implicit none + integer i + + !ERROR: The parameter of the SIZES clause must be a positive integer expression + !$omp tile sizes(0) + do i = 1, 5 + print *, i + end do +end subroutine + + +subroutine sizes_negative + implicit none + integer i + + !ERROR: The parameter of the SIZES clause must be a positive integer expression + !$omp tile sizes(-1) + do i = 1, 5 + print *, i + end do +end subroutine diff --git a/flang/test/Semantics/OpenMP/tile05.f90 b/flang/test/Semantics/OpenMP/tile05.f90 new file mode 100644 index 0000000000000..70c43811a5832 --- /dev/null +++ b/flang/test/Semantics/OpenMP/tile05.f90 @@ -0,0 +1,14 @@ +! Testing the Semantics of tile +!RUN: %python %S/../test_errors.py %s %flang -fopenmp -fopenmp-version=51 + + +subroutine insufficient_loops + implicit none + integer i + + !ERROR: The SIZES clause has more entries than there are nested canonical loops. + !$omp tile sizes(2, 2) + do i = 1, 5 + print *, i + end do +end subroutine diff --git a/flang/test/Semantics/OpenMP/tile06.f90 b/flang/test/Semantics/OpenMP/tile06.f90 new file mode 100644 index 0000000000000..52518d43f0554 --- /dev/null +++ b/flang/test/Semantics/OpenMP/tile06.f90 @@ -0,0 +1,44 @@ +! Testing the Semantics of tile +!RUN: %python %S/../test_errors.py %s %flang -fopenmp -fopenmp-version=51 + + +subroutine nonrectangular_loop_lb + implicit none + integer i, j + + !ERROR: Trip count must be computable and invariant + !$omp tile sizes(2,2) + do i = 1, 5 + do j = 1, i + print *, i, j + end do + end do +end subroutine + + +subroutine nonrectangular_loop_ub + implicit none + integer i, j + + !ERROR: Trip count must be computable and invariant + !$omp tile sizes(2,2) + do i = 1, 5 + do j = 1, i + print *, i, j + end do + end do +end subroutine + + +subroutine nonrectangular_loop_step + implicit none + integer i, j + + !ERROR: Trip count must be computable and invariant + !$omp tile sizes(2,2) + do i = 1, 5 + do j = 1, 42, i + print *, i, j + end do + end do +end subroutine diff --git a/flang/test/Semantics/OpenMP/tile07.f90 b/flang/test/Semantics/OpenMP/tile07.f90 new file mode 100644 index 0000000000000..70a6f5fc529a4 --- /dev/null +++ b/flang/test/Semantics/OpenMP/tile07.f90 @@ -0,0 +1,35 @@ +! Testing the Semantics of tile +!RUN: %python %S/../test_errors.py %s %flang -fopenmp -fopenmp-version=51 + + +subroutine non_perfectly_nested_loop_behind + implicit none + integer i, j + + !ERROR: Canonical loop nest must be perfectly nested. + !$omp tile sizes(2,2) + do i = 1, 5 + do j = 1, 42 + print *, j + end do + print *, i + end do +end subroutine + + +subroutine non_perfectly_nested_loop_before + implicit none + integer i, j + + !ERROR: The SIZES clause has more entries than there are nested canonical loops. + !$omp tile sizes(2,2) + do i = 1, 5 + print *, i + do j = 1, 42 + print *, j + end do + end do +end subroutine + + + diff --git a/flang/test/Semantics/OpenMP/tile08.f90 b/flang/test/Semantics/OpenMP/tile08.f90 new file mode 100644 index 0000000000000..f42805cb81b7d --- /dev/null +++ b/flang/test/Semantics/OpenMP/tile08.f90 @@ -0,0 +1,15 @@ +! Testing the Semantics of tile +!RUN: %python %S/../test_errors.py %s %flang -fopenmp -fopenmp-version=51 + + +subroutine do_concurrent + implicit none + integer i, j + + + !$omp tile sizes(2,2) + !ERROR: DO CONCURRENT loops cannot form part of a loop nest. + do concurrent (i = 1:42, j = 1:42) + print *, i, j + end do +end subroutine diff --git a/llvm/include/llvm/Frontend/OpenMP/OMP.td b/llvm/include/llvm/Frontend/OpenMP/OMP.td index 38f95a11bf85f..bba0d6ebdb791 100644 --- a/llvm/include/llvm/Frontend/OpenMP/OMP.td +++ b/llvm/include/llvm/Frontend/OpenMP/OMP.td @@ -1333,6 +1333,9 @@ def OMP_Tile : Directive<[Spelling<"tile">]> { let allowedOnceClauses = [ VersionedClause, ]; + let requiredClauses = [ + VersionedClause, + ]; let association = AS_Loop; let category = CA_Executable; } diff --git a/mlir/test/Target/LLVMIR/openmp-cli-tile01.mlir b/mlir/test/Target/LLVMIR/openmp-cli-tile01.mlir index 4ac4f02103e8c..0d559b69a3ad1 100644 --- a/mlir/test/Target/LLVMIR/openmp-cli-tile01.mlir +++ b/mlir/test/Target/LLVMIR/openmp-cli-tile01.mlir @@ -1,5 +1,4 @@ -// RUN: mlir-translate -mlir-to-llvmir %s | FileCheck %s - +// RUN: mlir-translate -mlir-to-llvmir %s | FileCheck %s --enable-var-scope llvm.func @tile_trivial_loop(%baseptr: !llvm.ptr, %tc: i32, %ts: i32) -> () { @@ -15,87 +14,81 @@ llvm.func @tile_trivial_loop(%baseptr: !llvm.ptr, %tc: i32, %ts: i32) -> () { } -// CHECK: ; ModuleID = 'LLVMDialectModule' -// CHECK-NEXT: source_filename = "LLVMDialectModule" -// CHECK-EMPTY: -// CHECK-NEXT: define void @tile_trivial_loop(ptr %0, i32 %1, i32 %2) { -// CHECK-NEXT: br label %omp_omp.loop.preheader -// CHECK-EMPTY: -// CHECK-NEXT: omp_omp.loop.preheader: ; preds = %3 -// CHECK-NEXT: %4 = udiv i32 %1, %2 -// CHECK-NEXT: %5 = urem i32 %1, %2 -// CHECK-NEXT: %6 = icmp ne i32 %5, 0 -// CHECK-NEXT: %7 = zext i1 %6 to i32 -// CHECK-NEXT: %omp_floor0.tripcount = add nuw i32 %4, %7 -// CHECK-NEXT: br label %omp_floor0.preheader -// CHECK-EMPTY: -// CHECK-NEXT: omp_floor0.preheader: ; preds = %omp_omp.loop.preheader -// CHECK-NEXT: br label %omp_floor0.header +// CHECK-LABEL: define void @tile_trivial_loop( +// CHECK-SAME: ptr %[[TMP0:.+]], i32 %[[TMP1:.+]], i32 %[[TMP2:.+]]) { +// CHECK-NEXT: br label %[[OMP_OMP_LOOP_PREHEADER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_floor0.header: ; preds = %omp_floor0.inc, %omp_floor0.preheader -// CHECK-NEXT: %omp_floor0.iv = phi i32 [ 0, %omp_floor0.preheader ], [ %omp_floor0.next, %omp_floor0.inc ] -// CHECK-NEXT: br label %omp_floor0.cond +// CHECK-NEXT: [[OMP_OMP_LOOP_PREHEADER]]: +// CHECK-NEXT: %[[TMP4:.+]] = udiv i32 %[[TMP1:.+]], %[[TMP2:.+]] +// CHECK-NEXT: %[[TMP5:.+]] = urem i32 %[[TMP1:.+]], %[[TMP2:.+]] +// CHECK-NEXT: %[[TMP6:.+]] = icmp ne i32 %[[TMP5:.+]], 0 +// CHECK-NEXT: %[[TMP7:.+]] = zext i1 %[[TMP6:.+]] to i32 +// CHECK-NEXT: %[[OMP_FLOOR0_TRIPCOUNT:.+]] = add nuw i32 %[[TMP4:.+]], %[[TMP7:.+]] +// CHECK-NEXT: br label %[[OMP_FLOOR0_PREHEADER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_floor0.cond: ; preds = %omp_floor0.header -// CHECK-NEXT: %omp_floor0.cmp = icmp ult i32 %omp_floor0.iv, %omp_floor0.tripcount -// CHECK-NEXT: br i1 %omp_floor0.cmp, label %omp_floor0.body, label %omp_floor0.exit +// CHECK-NEXT: [[OMP_FLOOR0_PREHEADER]]: +// CHECK-NEXT: br label %[[OMP_FLOOR0_HEADER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_floor0.body: ; preds = %omp_floor0.cond -// CHECK-NEXT: %8 = icmp eq i32 %omp_floor0.iv, %4 -// CHECK-NEXT: %9 = select i1 %8, i32 %5, i32 %2 -// CHECK-NEXT: br label %omp_tile0.preheader +// CHECK-NEXT: [[OMP_FLOOR0_HEADER]]: +// CHECK-NEXT: %[[OMP_FLOOR0_IV:.+]] = phi i32 [ 0, %[[OMP_FLOOR0_PREHEADER:.+]] ], [ %[[OMP_FLOOR0_NEXT:.+]], %[[OMP_FLOOR0_INC:.+]] ] +// CHECK-NEXT: br label %[[OMP_FLOOR0_COND:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_tile0.preheader: ; preds = %omp_floor0.body -// CHECK-NEXT: br label %omp_tile0.header +// CHECK-NEXT: [[OMP_FLOOR0_COND]]: +// CHECK-NEXT: %[[OMP_FLOOR0_CMP:.+]] = icmp ult i32 %[[OMP_FLOOR0_IV:.+]], %[[OMP_FLOOR0_TRIPCOUNT:.+]] +// CHECK-NEXT: br i1 %[[OMP_FLOOR0_CMP:.+]], label %[[OMP_FLOOR0_BODY:.+]], label %[[OMP_FLOOR0_EXIT:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_tile0.header: ; preds = %omp_tile0.inc, %omp_tile0.preheader -// CHECK-NEXT: %omp_tile0.iv = phi i32 [ 0, %omp_tile0.preheader ], [ %omp_tile0.next, %omp_tile0.inc ] -// CHECK-NEXT: br label %omp_tile0.cond +// CHECK-NEXT: [[OMP_FLOOR0_BODY]]: +// CHECK-NEXT: %[[TMP8:.+]] = icmp eq i32 %[[OMP_FLOOR0_IV:.+]], %[[TMP4:.+]] +// CHECK-NEXT: %[[TMP9:.+]] = select i1 %[[TMP8:.+]], i32 %[[TMP5:.+]], i32 %[[TMP2:.+]] +// CHECK-NEXT: br label %[[OMP_TILE0_PREHEADER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_tile0.cond: ; preds = %omp_tile0.header -// CHECK-NEXT: %omp_tile0.cmp = icmp ult i32 %omp_tile0.iv, %9 -// CHECK-NEXT: br i1 %omp_tile0.cmp, label %omp_tile0.body, label %omp_tile0.exit +// CHECK-NEXT: [[OMP_TILE0_PREHEADER]]: +// CHECK-NEXT: br label %[[OMP_TILE0_HEADER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_tile0.body: ; preds = %omp_tile0.cond -// CHECK-NEXT: %10 = mul nuw i32 %2, %omp_floor0.iv -// CHECK-NEXT: %11 = add nuw i32 %10, %omp_tile0.iv -// CHECK-NEXT: br label %omp_omp.loop.body +// CHECK-NEXT: [[OMP_TILE0_HEADER]]: +// CHECK-NEXT: %[[OMP_TILE0_IV:.+]] = phi i32 [ 0, %[[OMP_TILE0_PREHEADER:.+]] ], [ %[[OMP_TILE0_NEXT:.+]], %[[OMP_TILE0_INC:.+]] ] +// CHECK-NEXT: br label %[[OMP_TILE0_COND:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_omp.loop.body: ; preds = %omp_tile0.body -// CHECK-NEXT: br label %omp.loop.region +// CHECK-NEXT: [[OMP_TILE0_COND]]: +// CHECK-NEXT: %[[OMP_TILE0_CMP:.+]] = icmp ult i32 %[[OMP_TILE0_IV:.+]], %[[TMP9:.+]] +// CHECK-NEXT: br i1 %[[OMP_TILE0_CMP:.+]], label %[[OMP_TILE0_BODY:.+]], label %[[OMP_TILE0_EXIT:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp.loop.region: ; preds = %omp_omp.loop.body -// CHECK-NEXT: %12 = getelementptr inbounds float, ptr %0, i32 %11 -// CHECK-NEXT: store float 4.200000e+01, ptr %12, align 4 -// CHECK-NEXT: br label %omp.region.cont +// CHECK-NEXT: [[OMP_TILE0_BODY]]: +// CHECK-NEXT: %[[TMP10:.+]] = mul nuw i32 %[[TMP2:.+]], %[[OMP_FLOOR0_IV:.+]] +// CHECK-NEXT: %[[TMP11:.+]] = add nuw i32 %[[TMP10:.+]], %[[OMP_TILE0_IV:.+]] +// CHECK-NEXT: br label %[[OMP_OMP_LOOP_BODY:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp.region.cont: ; preds = %omp.loop.region -// CHECK-NEXT: br label %omp_tile0.inc +// CHECK-NEXT: [[OMP_OMP_LOOP_BODY]]: +// CHECK-NEXT: br label %[[OMP_LOOP_REGION:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_tile0.inc: ; preds = %omp.region.cont -// CHECK-NEXT: %omp_tile0.next = add nuw i32 %omp_tile0.iv, 1 -// CHECK-NEXT: br label %omp_tile0.header +// CHECK-NEXT: [[OMP_LOOP_REGION]]: +// CHECK-NEXT: %[[TMP12:.+]] = getelementptr inbounds float, ptr %[[TMP0:.+]], i32 %[[TMP11:.+]] +// CHECK-NEXT: store float 4.200000e+01, ptr %[[TMP12:.+]], align 4 +// CHECK-NEXT: br label %[[OMP_REGION_CONT:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_tile0.exit: ; preds = %omp_tile0.cond -// CHECK-NEXT: br label %omp_tile0.after +// CHECK-NEXT: [[OMP_REGION_CONT]]: +// CHECK-NEXT: br label %[[OMP_TILE0_INC:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_tile0.after: ; preds = %omp_tile0.exit -// CHECK-NEXT: br label %omp_floor0.inc +// CHECK-NEXT: [[OMP_TILE0_INC]]: +// CHECK-NEXT: %[[OMP_TILE0_NEXT:.+]] = add nuw i32 %[[OMP_TILE0_IV:.+]], 1 +// CHECK-NEXT: br label %[[OMP_TILE0_HEADER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_floor0.inc: ; preds = %omp_tile0.after -// CHECK-NEXT: %omp_floor0.next = add nuw i32 %omp_floor0.iv, 1 -// CHECK-NEXT: br label %omp_floor0.header +// CHECK-NEXT: [[OMP_TILE0_EXIT]]: +// CHECK-NEXT: br label %[[OMP_TILE0_AFTER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_floor0.exit: ; preds = %omp_floor0.cond -// CHECK-NEXT: br label %omp_floor0.after +// CHECK-NEXT: [[OMP_TILE0_AFTER]]: +// CHECK-NEXT: br label %[[OMP_FLOOR0_INC:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_floor0.after: ; preds = %omp_floor0.exit -// CHECK-NEXT: br label %omp_omp.loop.after +// CHECK-NEXT: [[OMP_FLOOR0_INC]]: +// CHECK-NEXT: %[[OMP_FLOOR0_NEXT:.+]] = add nuw i32 %[[OMP_FLOOR0_IV:.+]], 1 +// CHECK-NEXT: br label %[[OMP_FLOOR0_HEADER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_omp.loop.after: ; preds = %omp_floor0.after -// CHECK-NEXT: ret void -// CHECK-NEXT: } +// CHECK-NEXT: [[OMP_FLOOR0_EXIT]]: +// CHECK-NEXT: br label %[[OMP_FLOOR0_AFTER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: !llvm.module.flags = !{!0} +// CHECK-NEXT: [[OMP_FLOOR0_AFTER]]: +// CHECK-NEXT: br label %[[OMP_OMP_LOOP_AFTER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: !0 = !{i32 2, !"Debug Info Version", i32 3} +// CHECK-NEXT: [[OMP_OMP_LOOP_AFTER]]: +// CHECK-NEXT: ret void +// CHECK-NEXT: } diff --git a/mlir/test/Target/LLVMIR/openmp-cli-tile02.mlir b/mlir/test/Target/LLVMIR/openmp-cli-tile02.mlir index 6fad81cd0c299..22c2973164159 100644 --- a/mlir/test/Target/LLVMIR/openmp-cli-tile02.mlir +++ b/mlir/test/Target/LLVMIR/openmp-cli-tile02.mlir @@ -1,4 +1,4 @@ -// RUN: mlir-translate -mlir-to-llvmir %s | FileCheck %s +// RUN: mlir-translate -mlir-to-llvmir %s | FileCheck %s --enable-var-scope llvm.func @tile_2d_loop(%baseptr: !llvm.ptr, %tc1: i32, %tc2: i32, %ts1: i32, %ts2: i32) -> () { @@ -19,172 +19,166 @@ llvm.func @tile_2d_loop(%baseptr: !llvm.ptr, %tc1: i32, %tc2: i32, %ts1: i32, %t } -// CHECK: ; ModuleID = 'LLVMDialectModule' -// CHECK-NEXT: source_filename = "LLVMDialectModule" +// CHECK-LABEL: define void @tile_2d_loop( +// CHECK-SAME: ptr %[[TMP0:.+]], i32 %[[TMP1:.+]], i32 %[[TMP2:.+]], i32 %[[TMP3:.+]], i32 %[[TMP4:.+]]) { +// CHECK-NEXT: br label %[[OMP_OMP_LOOP_PREHEADER:.+]] +// CHECK-EMPTY: +// CHECK-NEXT: [[OMP_OMP_LOOP_PREHEADER]]: +// CHECK-NEXT: %[[TMP6:.+]] = udiv i32 %[[TMP1:.+]], %[[TMP3:.+]] +// CHECK-NEXT: %[[TMP7:.+]] = urem i32 %[[TMP1:.+]], %[[TMP3:.+]] +// CHECK-NEXT: %[[TMP8:.+]] = icmp ne i32 %[[TMP7:.+]], 0 +// CHECK-NEXT: %[[TMP9:.+]] = zext i1 %[[TMP8:.+]] to i32 +// CHECK-NEXT: %[[OMP_FLOOR0_TRIPCOUNT:.+]] = add nuw i32 %[[TMP6:.+]], %[[TMP9:.+]] +// CHECK-NEXT: %[[TMP10:.+]] = udiv i32 %[[TMP2:.+]], %[[TMP4:.+]] +// CHECK-NEXT: %[[TMP11:.+]] = urem i32 %[[TMP2:.+]], %[[TMP4:.+]] +// CHECK-NEXT: %[[TMP12:.+]] = icmp ne i32 %[[TMP11:.+]], 0 +// CHECK-NEXT: %[[TMP13:.+]] = zext i1 %[[TMP12:.+]] to i32 +// CHECK-NEXT: %[[OMP_FLOOR1_TRIPCOUNT:.+]] = add nuw i32 %[[TMP10:.+]], %[[TMP13:.+]] +// CHECK-NEXT: br label %[[OMP_FLOOR0_PREHEADER:.+]] +// CHECK-EMPTY: +// CHECK-NEXT: [[OMP_OMP_LOOP_HEADER:.+]]: +// CHECK-NEXT: %[[OMP_OMP_LOOP_IV:.+]] = phi i32 [ %[[OMP_OMP_LOOP_NEXT:.+]], %[[OMP_OMP_LOOP_INC:.+]] ] +// CHECK-NEXT: br label %[[OMP_OMP_LOOP_COND:.+]] +// CHECK-EMPTY: +// CHECK-NEXT: [[OMP_OMP_LOOP_COND]]: +// CHECK-NEXT: %[[OMP_OMP_LOOP_CMP:.+]] = icmp ult i32 %[[TMP19:.+]], %[[TMP1:.+]] +// CHECK-NEXT: br i1 %[[OMP_OMP_LOOP_CMP:.+]], label %[[OMP_OMP_LOOP_BODY:.+]], label %[[OMP_OMP_LOOP_EXIT:.+]] +// CHECK-EMPTY: +// CHECK-NEXT: [[OMP_OMP_LOOP_BODY]]: +// CHECK-NEXT: br label %[[OMP_LOOP_REGION:.+]] +// CHECK-EMPTY: +// CHECK-NEXT: [[OMP_LOOP_REGION]]: +// CHECK-NEXT: br label %[[OMP_OMP_LOOP_PREHEADER1:.+]] // CHECK-EMPTY: -// CHECK-NEXT: define void @tile_2d_loop(ptr %0, i32 %1, i32 %2, i32 %3, i32 %4) { -// CHECK-NEXT: br label %omp_omp.loop.preheader -// CHECK-EMPTY: -// CHECK-NEXT: omp_omp.loop.preheader: ; preds = %5 -// CHECK-NEXT: %6 = udiv i32 %1, %3 -// CHECK-NEXT: %7 = urem i32 %1, %3 -// CHECK-NEXT: %8 = icmp ne i32 %7, 0 -// CHECK-NEXT: %9 = zext i1 %8 to i32 -// CHECK-NEXT: %omp_floor0.tripcount = add nuw i32 %6, %9 -// CHECK-NEXT: %10 = udiv i32 %2, %4 -// CHECK-NEXT: %11 = urem i32 %2, %4 -// CHECK-NEXT: %12 = icmp ne i32 %11, 0 -// CHECK-NEXT: %13 = zext i1 %12 to i32 -// CHECK-NEXT: %omp_floor1.tripcount = add nuw i32 %10, %13 -// CHECK-NEXT: br label %omp_floor0.preheader +// CHECK-NEXT: [[OMP_OMP_LOOP_PREHEADER1]]: +// CHECK-NEXT: br label %[[OMP_OMP_LOOP_BODY4:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_omp.loop.header: ; preds = %omp_omp.loop.inc -// CHECK-NEXT: %omp_omp.loop.iv = phi i32 [ %omp_omp.loop.next, %omp_omp.loop.inc ] -// CHECK-NEXT: br label %omp_omp.loop.cond +// CHECK-NEXT: [[OMP_FLOOR0_PREHEADER]]: +// CHECK-NEXT: br label %[[OMP_FLOOR0_HEADER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_omp.loop.cond: ; preds = %omp_omp.loop.header -// CHECK-NEXT: %omp_omp.loop.cmp = icmp ult i32 %19, %1 -// CHECK-NEXT: br i1 %omp_omp.loop.cmp, label %omp_omp.loop.body, label %omp_omp.loop.exit +// CHECK-NEXT: [[OMP_FLOOR0_HEADER]]: +// CHECK-NEXT: %[[OMP_FLOOR0_IV:.+]] = phi i32 [ 0, %[[OMP_FLOOR0_PREHEADER:.+]] ], [ %[[OMP_FLOOR0_NEXT:.+]], %[[OMP_FLOOR0_INC:.+]] ] +// CHECK-NEXT: br label %[[OMP_FLOOR0_COND:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_omp.loop.body: ; preds = %omp_tile1.body, %omp_omp.loop.cond -// CHECK-NEXT: br label %omp.loop.region +// CHECK-NEXT: [[OMP_FLOOR0_COND]]: +// CHECK-NEXT: %[[OMP_FLOOR0_CMP:.+]] = icmp ult i32 %[[OMP_FLOOR0_IV:.+]], %[[OMP_FLOOR0_TRIPCOUNT:.+]] +// CHECK-NEXT: br i1 %[[OMP_FLOOR0_CMP:.+]], label %[[OMP_FLOOR0_BODY:.+]], label %[[OMP_FLOOR0_EXIT:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp.loop.region: ; preds = %omp_omp.loop.body -// CHECK-NEXT: br label %omp_omp.loop.preheader1 +// CHECK-NEXT: [[OMP_FLOOR0_BODY]]: +// CHECK-NEXT: br label %[[OMP_FLOOR1_PREHEADER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_omp.loop.preheader1: ; preds = %omp.loop.region -// CHECK-NEXT: br label %omp_omp.loop.body4 +// CHECK-NEXT: [[OMP_FLOOR1_PREHEADER]]: +// CHECK-NEXT: br label %[[OMP_FLOOR1_HEADER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_floor0.preheader: ; preds = %omp_omp.loop.preheader -// CHECK-NEXT: br label %omp_floor0.header +// CHECK-NEXT: [[OMP_FLOOR1_HEADER]]: +// CHECK-NEXT: %[[OMP_FLOOR1_IV:.+]] = phi i32 [ 0, %[[OMP_FLOOR1_PREHEADER:.+]] ], [ %[[OMP_FLOOR1_NEXT:.+]], %[[OMP_FLOOR1_INC:.+]] ] +// CHECK-NEXT: br label %[[OMP_FLOOR1_COND:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_floor0.header: ; preds = %omp_floor0.inc, %omp_floor0.preheader -// CHECK-NEXT: %omp_floor0.iv = phi i32 [ 0, %omp_floor0.preheader ], [ %omp_floor0.next, %omp_floor0.inc ] -// CHECK-NEXT: br label %omp_floor0.cond +// CHECK-NEXT: [[OMP_FLOOR1_COND]]: +// CHECK-NEXT: %[[OMP_FLOOR1_CMP:.+]] = icmp ult i32 %[[OMP_FLOOR1_IV:.+]], %[[OMP_FLOOR1_TRIPCOUNT:.+]] +// CHECK-NEXT: br i1 %[[OMP_FLOOR1_CMP:.+]], label %[[OMP_FLOOR1_BODY:.+]], label %[[OMP_FLOOR1_EXIT:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_floor0.cond: ; preds = %omp_floor0.header -// CHECK-NEXT: %omp_floor0.cmp = icmp ult i32 %omp_floor0.iv, %omp_floor0.tripcount -// CHECK-NEXT: br i1 %omp_floor0.cmp, label %omp_floor0.body, label %omp_floor0.exit +// CHECK-NEXT: [[OMP_FLOOR1_BODY]]: +// CHECK-NEXT: %[[TMP14:.+]] = icmp eq i32 %[[OMP_FLOOR0_IV:.+]], %[[TMP6:.+]] +// CHECK-NEXT: %[[TMP15:.+]] = select i1 %[[TMP14:.+]], i32 %[[TMP7:.+]], i32 %[[TMP3:.+]] +// CHECK-NEXT: %[[TMP16:.+]] = icmp eq i32 %[[OMP_FLOOR1_IV:.+]], %[[TMP10:.+]] +// CHECK-NEXT: %[[TMP17:.+]] = select i1 %[[TMP16:.+]], i32 %[[TMP11:.+]], i32 %[[TMP4:.+]] +// CHECK-NEXT: br label %[[OMP_TILE0_PREHEADER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_floor0.body: ; preds = %omp_floor0.cond -// CHECK-NEXT: br label %omp_floor1.preheader +// CHECK-NEXT: [[OMP_TILE0_PREHEADER]]: +// CHECK-NEXT: br label %[[OMP_TILE0_HEADER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_floor1.preheader: ; preds = %omp_floor0.body -// CHECK-NEXT: br label %omp_floor1.header +// CHECK-NEXT: [[OMP_TILE0_HEADER]]: +// CHECK-NEXT: %[[OMP_TILE0_IV:.+]] = phi i32 [ 0, %[[OMP_TILE0_PREHEADER:.+]] ], [ %[[OMP_TILE0_NEXT:.+]], %[[OMP_TILE0_INC:.+]] ] +// CHECK-NEXT: br label %[[OMP_TILE0_COND:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_floor1.header: ; preds = %omp_floor1.inc, %omp_floor1.preheader -// CHECK-NEXT: %omp_floor1.iv = phi i32 [ 0, %omp_floor1.preheader ], [ %omp_floor1.next, %omp_floor1.inc ] -// CHECK-NEXT: br label %omp_floor1.cond +// CHECK-NEXT: [[OMP_TILE0_COND]]: +// CHECK-NEXT: %[[OMP_TILE0_CMP:.+]] = icmp ult i32 %[[OMP_TILE0_IV:.+]], %[[TMP15:.+]] +// CHECK-NEXT: br i1 %[[OMP_TILE0_CMP:.+]], label %[[OMP_TILE0_BODY:.+]], label %[[OMP_TILE0_EXIT:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_floor1.cond: ; preds = %omp_floor1.header -// CHECK-NEXT: %omp_floor1.cmp = icmp ult i32 %omp_floor1.iv, %omp_floor1.tripcount -// CHECK-NEXT: br i1 %omp_floor1.cmp, label %omp_floor1.body, label %omp_floor1.exit +// CHECK-NEXT: [[OMP_TILE0_BODY]]: +// CHECK-NEXT: br label %[[OMP_TILE1_PREHEADER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_floor1.body: ; preds = %omp_floor1.cond -// CHECK-NEXT: %14 = icmp eq i32 %omp_floor0.iv, %6 -// CHECK-NEXT: %15 = select i1 %14, i32 %7, i32 %3 -// CHECK-NEXT: %16 = icmp eq i32 %omp_floor1.iv, %10 -// CHECK-NEXT: %17 = select i1 %16, i32 %11, i32 %4 -// CHECK-NEXT: br label %omp_tile0.preheader +// CHECK-NEXT: [[OMP_TILE1_PREHEADER]]: +// CHECK-NEXT: br label %[[OMP_TILE1_HEADER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_tile0.preheader: ; preds = %omp_floor1.body -// CHECK-NEXT: br label %omp_tile0.header +// CHECK-NEXT: [[OMP_TILE1_HEADER]]: +// CHECK-NEXT: %[[OMP_TILE1_IV:.+]] = phi i32 [ 0, %[[OMP_TILE1_PREHEADER:.+]] ], [ %[[OMP_TILE1_NEXT:.+]], %[[OMP_TILE1_INC:.+]] ] +// CHECK-NEXT: br label %[[OMP_TILE1_COND:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_tile0.header: ; preds = %omp_tile0.inc, %omp_tile0.preheader -// CHECK-NEXT: %omp_tile0.iv = phi i32 [ 0, %omp_tile0.preheader ], [ %omp_tile0.next, %omp_tile0.inc ] -// CHECK-NEXT: br label %omp_tile0.cond +// CHECK-NEXT: [[OMP_TILE1_COND]]: +// CHECK-NEXT: %[[OMP_TILE1_CMP:.+]] = icmp ult i32 %[[OMP_TILE1_IV:.+]], %[[TMP17:.+]] +// CHECK-NEXT: br i1 %[[OMP_TILE1_CMP:.+]], label %[[OMP_TILE1_BODY:.+]], label %[[OMP_TILE1_EXIT:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_tile0.cond: ; preds = %omp_tile0.header -// CHECK-NEXT: %omp_tile0.cmp = icmp ult i32 %omp_tile0.iv, %15 -// CHECK-NEXT: br i1 %omp_tile0.cmp, label %omp_tile0.body, label %omp_tile0.exit +// CHECK-NEXT: [[OMP_TILE1_BODY]]: +// CHECK-NEXT: %[[TMP18:.+]] = mul nuw i32 %[[TMP3:.+]], %[[OMP_FLOOR0_IV:.+]] +// CHECK-NEXT: %[[TMP19:.+]] = add nuw i32 %[[TMP18:.+]], %[[OMP_TILE0_IV:.+]] +// CHECK-NEXT: %[[TMP20:.+]] = mul nuw i32 %[[TMP4:.+]], %[[OMP_FLOOR1_IV:.+]] +// CHECK-NEXT: %[[TMP21:.+]] = add nuw i32 %[[TMP20:.+]], %[[OMP_TILE1_IV:.+]] +// CHECK-NEXT: br label %[[OMP_OMP_LOOP_BODY:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_tile0.body: ; preds = %omp_tile0.cond -// CHECK-NEXT: br label %omp_tile1.preheader +// CHECK-NEXT: [[OMP_OMP_LOOP_BODY4]]: +// CHECK-NEXT: br label %[[OMP_LOOP_REGION12:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_tile1.preheader: ; preds = %omp_tile0.body -// CHECK-NEXT: br label %omp_tile1.header +// CHECK-NEXT: [[OMP_LOOP_REGION12]]: +// CHECK-NEXT: %[[TMP22:.+]] = add i32 %[[TMP19:.+]], %[[TMP21:.+]] +// CHECK-NEXT: %[[TMP23:.+]] = getelementptr inbounds float, ptr %[[TMP0:.+]], i32 %[[TMP22:.+]] +// CHECK-NEXT: store float 4.200000e+01, ptr %[[TMP23:.+]], align 4 +// CHECK-NEXT: br label %[[OMP_REGION_CONT11:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_tile1.header: ; preds = %omp_tile1.inc, %omp_tile1.preheader -// CHECK-NEXT: %omp_tile1.iv = phi i32 [ 0, %omp_tile1.preheader ], [ %omp_tile1.next, %omp_tile1.inc ] -// CHECK-NEXT: br label %omp_tile1.cond +// CHECK-NEXT: [[OMP_REGION_CONT11]]: +// CHECK-NEXT: br label %[[OMP_TILE1_INC:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_tile1.cond: ; preds = %omp_tile1.header -// CHECK-NEXT: %omp_tile1.cmp = icmp ult i32 %omp_tile1.iv, %17 -// CHECK-NEXT: br i1 %omp_tile1.cmp, label %omp_tile1.body, label %omp_tile1.exit +// CHECK-NEXT: [[OMP_TILE1_INC]]: +// CHECK-NEXT: %[[OMP_TILE1_NEXT:.+]] = add nuw i32 %[[OMP_TILE1_IV:.+]], 1 +// CHECK-NEXT: br label %[[OMP_TILE1_HEADER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_tile1.body: ; preds = %omp_tile1.cond -// CHECK-NEXT: %18 = mul nuw i32 %3, %omp_floor0.iv -// CHECK-NEXT: %19 = add nuw i32 %18, %omp_tile0.iv -// CHECK-NEXT: %20 = mul nuw i32 %4, %omp_floor1.iv -// CHECK-NEXT: %21 = add nuw i32 %20, %omp_tile1.iv -// CHECK-NEXT: br label %omp_omp.loop.body +// CHECK-NEXT: [[OMP_TILE1_EXIT]]: +// CHECK-NEXT: br label %[[OMP_TILE1_AFTER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_omp.loop.body4: ; preds = %omp_omp.loop.preheader1 -// CHECK-NEXT: br label %omp.loop.region12 +// CHECK-NEXT: [[OMP_TILE1_AFTER]]: +// CHECK-NEXT: br label %[[OMP_TILE0_INC:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp.loop.region12: ; preds = %omp_omp.loop.body4 -// CHECK-NEXT: %22 = add i32 %19, %21 -// CHECK-NEXT: %23 = getelementptr inbounds float, ptr %0, i32 %22 -// CHECK-NEXT: store float 4.200000e+01, ptr %23, align 4 -// CHECK-NEXT: br label %omp.region.cont11 +// CHECK-NEXT: [[OMP_TILE0_INC]]: +// CHECK-NEXT: %[[OMP_TILE0_NEXT:.+]] = add nuw i32 %[[OMP_TILE0_IV:.+]], 1 +// CHECK-NEXT: br label %[[OMP_TILE0_HEADER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp.region.cont11: ; preds = %omp.loop.region12 -// CHECK-NEXT: br label %omp_tile1.inc +// CHECK-NEXT: [[OMP_TILE0_EXIT]]: +// CHECK-NEXT: br label %[[OMP_TILE0_AFTER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_tile1.inc: ; preds = %omp.region.cont11 -// CHECK-NEXT: %omp_tile1.next = add nuw i32 %omp_tile1.iv, 1 -// CHECK-NEXT: br label %omp_tile1.header +// CHECK-NEXT: [[OMP_TILE0_AFTER]]: +// CHECK-NEXT: br label %[[OMP_FLOOR1_INC:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_tile1.exit: ; preds = %omp_tile1.cond -// CHECK-NEXT: br label %omp_tile1.after +// CHECK-NEXT: [[OMP_FLOOR1_INC]]: +// CHECK-NEXT: %[[OMP_FLOOR1_NEXT:.+]] = add nuw i32 %[[OMP_FLOOR1_IV:.+]], 1 +// CHECK-NEXT: br label %[[OMP_FLOOR1_HEADER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_tile1.after: ; preds = %omp_tile1.exit -// CHECK-NEXT: br label %omp_tile0.inc +// CHECK-NEXT: [[OMP_FLOOR1_EXIT]]: +// CHECK-NEXT: br label %[[OMP_FLOOR1_AFTER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_tile0.inc: ; preds = %omp_tile1.after -// CHECK-NEXT: %omp_tile0.next = add nuw i32 %omp_tile0.iv, 1 -// CHECK-NEXT: br label %omp_tile0.header +// CHECK-NEXT: [[OMP_FLOOR1_AFTER]]: +// CHECK-NEXT: br label %[[OMP_FLOOR0_INC:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_tile0.exit: ; preds = %omp_tile0.cond -// CHECK-NEXT: br label %omp_tile0.after +// CHECK-NEXT: [[OMP_FLOOR0_INC]]: +// CHECK-NEXT: %[[OMP_FLOOR0_NEXT:.+]] = add nuw i32 %[[OMP_FLOOR0_IV:.+]], 1 +// CHECK-NEXT: br label %[[OMP_FLOOR0_HEADER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_tile0.after: ; preds = %omp_tile0.exit -// CHECK-NEXT: br label %omp_floor1.inc +// CHECK-NEXT: [[OMP_FLOOR0_EXIT]]: +// CHECK-NEXT: br label %[[OMP_FLOOR0_AFTER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_floor1.inc: ; preds = %omp_tile0.after -// CHECK-NEXT: %omp_floor1.next = add nuw i32 %omp_floor1.iv, 1 -// CHECK-NEXT: br label %omp_floor1.header +// CHECK-NEXT: [[OMP_FLOOR0_AFTER]]: +// CHECK-NEXT: br label %[[OMP_OMP_LOOP_AFTER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_floor1.exit: ; preds = %omp_floor1.cond -// CHECK-NEXT: br label %omp_floor1.after +// CHECK-NEXT: [[OMP_REGION_CONT:.+]]: +// CHECK-NEXT: br label %[[OMP_OMP_LOOP_INC:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_floor1.after: ; preds = %omp_floor1.exit -// CHECK-NEXT: br label %omp_floor0.inc +// CHECK-NEXT: [[OMP_OMP_LOOP_INC]]: +// CHECK-NEXT: %[[OMP_OMP_LOOP_NEXT:.+]] = add nuw i32 %[[TMP19:.+]], 1 +// CHECK-NEXT: br label %[[OMP_OMP_LOOP_HEADER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_floor0.inc: ; preds = %omp_floor1.after -// CHECK-NEXT: %omp_floor0.next = add nuw i32 %omp_floor0.iv, 1 -// CHECK-NEXT: br label %omp_floor0.header +// CHECK-NEXT: [[OMP_OMP_LOOP_EXIT]]: +// CHECK-NEXT: br label %[[OMP_OMP_LOOP_AFTER:.+]] // CHECK-EMPTY: -// CHECK-NEXT: omp_floor0.exit: ; preds = %omp_floor0.cond -// CHECK-NEXT: br label %omp_floor0.after -// CHECK-EMPTY: -// CHECK-NEXT: omp_floor0.after: ; preds = %omp_floor0.exit -// CHECK-NEXT: br label %omp_omp.loop.after -// CHECK-EMPTY: -// CHECK-NEXT: omp.region.cont: ; No predecessors! -// CHECK-NEXT: br label %omp_omp.loop.inc -// CHECK-EMPTY: -// CHECK-NEXT: omp_omp.loop.inc: ; preds = %omp.region.cont -// CHECK-NEXT: %omp_omp.loop.next = add nuw i32 %19, 1 -// CHECK-NEXT: br label %omp_omp.loop.header -// CHECK-EMPTY: -// CHECK-NEXT: omp_omp.loop.exit: ; preds = %omp_omp.loop.cond -// CHECK-NEXT: br label %omp_omp.loop.after -// CHECK-EMPTY: -// CHECK-NEXT: omp_omp.loop.after: ; preds = %omp_floor0.after, %omp_omp.loop.exit -// CHECK-NEXT: ret void -// CHECK-NEXT: } -// CHECK-EMPTY: -// CHECK-NEXT: !llvm.module.flags = !{!0} -// CHECK-EMPTY: -// CHECK-NEXT: !0 = !{i32 2, !"Debug Info Version", i32 3} +// CHECK-NEXT: [[OMP_OMP_LOOP_AFTER]]: +// CHECK-NEXT: ret void +// CHECK-NEXT: } diff --git a/openmp/runtime/test/transform/tile/intfor.f90 b/openmp/runtime/test/transform/tile/intfor.f90 new file mode 100644 index 0000000000000..dac0de6a99021 --- /dev/null +++ b/openmp/runtime/test/transform/tile/intfor.f90 @@ -0,0 +1,31 @@ +! This test checks lowering of the OpenMP tile directive +! It is done 3 times corresponding to every possible fraction of the last +! iteration before passing beyond UB. + +! RUN: %flang %flags %openmp_flags -fopenmp-version=51 -DUB=16 %s -o %t-ub16.exe +! RUN: %flang %flags %openmp_flags -fopenmp-version=51 -DUB=17 %s -o %t-ub17.exe +! RUN: %flang %flags %openmp_flags -fopenmp-version=51 -DUB=18 %s -o %t-ub18.exe +! RUN: %t-ub16.exe | FileCheck %s --match-full-lines +! RUN: %t-ub17.exe | FileCheck %s --match-full-lines +! RUN: %t-ub18.exe | FileCheck %s --match-full-lines + +program tile_intfor_1d + integer i + print *, 'do' + + !$OMP TILE SIZES(2) + do i=7, UB, 3 + print '("i=", I0)', i + end do + !$OMP END TILE + + print *, 'done' +end program + + +! CHECK: do +! CHECK-NEXT: i=7 +! CHECK-NEXT: i=10 +! CHECK-NEXT: i=13 +! CHECK-NEXT: i=16 +! CHECK-NEXT: done diff --git a/openmp/runtime/test/transform/tile/intfor_2d.f90 b/openmp/runtime/test/transform/tile/intfor_2d.f90 new file mode 100644 index 0000000000000..6bc90c768b8d3 --- /dev/null +++ b/openmp/runtime/test/transform/tile/intfor_2d.f90 @@ -0,0 +1,53 @@ +! This test checks lowering of OpenMP tile directive + +! RUN: %flang %flags %openmp_flags -fopenmp-version=51 %s -o %t.exe +! RUN: %t.exe | FileCheck %s --match-full-lines + + +program tile_intfor_2d + integer i, j + print *, 'do' + + !$OMP TILE SIZES(2,3) + do i = 7, 16, 3 + do j = 0, 4 + print '("i=", I0," j=", I0)', i, j + end do + end do + !$OMP END TILE + + print *, 'done' +end program + + +! CHECK: do + +! complete tile +! CHECK-NEXT: i=7 j=0 +! CHECK-NEXT: i=7 j=1 +! CHECK-NEXT: i=7 j=2 +! CHECK-NEXT: i=10 j=0 +! CHECK-NEXT: i=10 j=1 +! CHECK-NEXT: i=10 j=2 + +! partial tile +! CHECK-NEXT: i=7 j=3 +! CHECK-NEXT: i=7 j=4 +! CHECK-NEXT: i=10 j=3 +! CHECK-NEXT: i=10 j=4 + +! complete tile +! CHECK-NEXT: i=13 j=0 +! CHECK-NEXT: i=13 j=1 +! CHECK-NEXT: i=13 j=2 +! CHECK-NEXT: i=16 j=0 +! CHECK-NEXT: i=16 j=1 +! CHECK-NEXT: i=16 j=2 + +! partial tile +! CHECK-NEXT: i=13 j=3 +! CHECK-NEXT: i=13 j=4 +! CHECK-NEXT: i=16 j=3 +! CHECK-NEXT: i=16 j=4 + +! CHECK-NEXT: done diff --git a/openmp/runtime/test/transform/tile/intfor_2d_varsizes.F90 b/openmp/runtime/test/transform/tile/intfor_2d_varsizes.F90 new file mode 100644 index 0000000000000..4cb5adf606dd2 --- /dev/null +++ b/openmp/runtime/test/transform/tile/intfor_2d_varsizes.F90 @@ -0,0 +1,60 @@ +! This test checks lowering of OpenMP tile directive + +! RUN: %flang %flags %openmp_flags -fopenmp-version=51 %s -o %t.exe +! RUN: %t.exe | FileCheck %s --match-full-lines + +program tile_intfor_varsizes + integer i + + call kernel(7,17,3,2) + call kernel(7,17,3,3) + +end program + + +subroutine kernel(lb, ub, step, ts) + integer i, j, lb, ub, step, ts + + print *, 'do' + + !$OMP TILE SIZES(ts,ts) + do i = lb, ub, step + do j = 0, 2 + print '("i=", I0," j=", I0)', i, j + end do + end do + !$OMP END TILE + + print *, 'done' + +end subroutine + +! CHECK: do +! CHECK-NEXT: i=7 j=0 +! CHECK-NEXT: i=7 j=1 +! CHECK-NEXT: i=10 j=0 +! CHECK-NEXT: i=10 j=1 +! CHECK-NEXT: i=7 j=2 +! CHECK-NEXT: i=10 j=2 +! CHECK-NEXT: i=13 j=0 +! CHECK-NEXT: i=13 j=1 +! CHECK-NEXT: i=16 j=0 +! CHECK-NEXT: i=16 j=1 +! CHECK-NEXT: i=13 j=2 +! CHECK-NEXT: i=16 j=2 +! CHECK-NEXT: done + +! CHECK: do +! CHECK-NEXT: i=7 j=0 +! CHECK-NEXT: i=7 j=1 +! CHECK-NEXT: i=7 j=2 +! CHECK-NEXT: i=10 j=0 +! CHECK-NEXT: i=10 j=1 +! CHECK-NEXT: i=10 j=2 +! CHECK-NEXT: i=13 j=0 +! CHECK-NEXT: i=13 j=1 +! CHECK-NEXT: i=13 j=2 +! CHECK-NEXT: i=16 j=0 +! CHECK-NEXT: i=16 j=1 +! CHECK-NEXT: i=16 j=2 +! CHECK-NEXT: done