-
Notifications
You must be signed in to change notification settings - Fork 14.3k
[Flang][OpenMP] Separate creation of work-sharing and SIMD loops, NFC #77757
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
@llvm/pr-subscribers-flang-fir-hlfir @llvm/pr-subscribers-flang-openmp Author: Krzysztof Parzyszek (kparzysz) ChangesThese two constructs were both handled in The shared code may be placed into a helper function later if needed. Recursive lowering [1/5] Full diff: https://github.com/llvm/llvm-project/pull/77757.diff 1 Files Affected:
diff --git a/flang/lib/Lower/OpenMP.cpp b/flang/lib/Lower/OpenMP.cpp
index c3a570bf15ea0d..99690b03eca1d3 100644
--- a/flang/lib/Lower/OpenMP.cpp
+++ b/flang/lib/Lower/OpenMP.cpp
@@ -2968,24 +2968,151 @@ genOMP(Fortran::lower::AbstractConverter &converter,
standaloneConstruct.u);
}
-static void genOMP(Fortran::lower::AbstractConverter &converter,
- Fortran::lower::pft::Evaluation &eval,
- Fortran::semantics::SemanticsContext &semanticsContext,
- const Fortran::parser::OpenMPLoopConstruct &loopConstruct) {
+static void
+createSimdLoop(Fortran::lower::AbstractConverter &converter,
+ Fortran::lower::pft::Evaluation &eval,
+ llvm::omp::Directive ompDirective,
+ const Fortran::parser::OmpClauseList &loopOpClauseList,
+ mlir::Location loc) {
fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
- llvm::SmallVector<mlir::Value> lowerBound, upperBound, step, linearVars,
- linearStepVars, reductionVars;
+ DataSharingProcessor dsp(converter, loopOpClauseList, eval);
+ dsp.processStep1();
+
+ Fortran::lower::StatementContext stmtCtx;
mlir::Value scheduleChunkClauseOperand;
- mlir::IntegerAttr orderedClauseOperand;
+ llvm::SmallVector<mlir::Value> lowerBound, upperBound, step, reductionVars;
+ llvm::SmallVector<const Fortran::semantics::Symbol *> iv;
+ llvm::SmallVector<mlir::Attribute> reductionDeclSymbols;
+ mlir::omp::ClauseOrderKindAttr orderClauseOperand;
+ std::size_t loopVarTypeSize;
+
+ ClauseProcessor cp(converter, loopOpClauseList);
+ cp.processCollapse(loc, eval, lowerBound, upperBound, step, iv,
+ loopVarTypeSize);
+ cp.processScheduleChunk(stmtCtx, scheduleChunkClauseOperand);
+ cp.processReduction(loc, reductionVars, reductionDeclSymbols);
+ cp.processTODO<Fortran::parser::OmpClause::Linear,
+ Fortran::parser::OmpClause::Order>(loc, ompDirective);
+
+ // The types of lower bound, upper bound, and step are converted into the
+ // type of the loop variable if necessary.
+ mlir::Type loopVarType = getLoopVarType(converter, loopVarTypeSize);
+ for (unsigned it = 0; it < (unsigned)lowerBound.size(); it++) {
+ lowerBound[it] =
+ firOpBuilder.createConvert(loc, loopVarType, lowerBound[it]);
+ upperBound[it] =
+ firOpBuilder.createConvert(loc, loopVarType, upperBound[it]);
+ step[it] = firOpBuilder.createConvert(loc, loopVarType, step[it]);
+ }
+
+ llvm::SmallVector<mlir::Value> alignedVars, nontemporalVars;
+ mlir::Value ifClauseOperand;
+ mlir::IntegerAttr simdlenClauseOperand, safelenClauseOperand;
+ cp.processIf(Fortran::parser::OmpIfClause::DirectiveNameModifier::Simd,
+ ifClauseOperand);
+ cp.processSimdlen(simdlenClauseOperand);
+ cp.processSafelen(safelenClauseOperand);
+ cp.processTODO<Fortran::parser::OmpClause::Aligned,
+ Fortran::parser::OmpClause::Allocate,
+ Fortran::parser::OmpClause::Nontemporal>(loc, ompDirective);
+
+ mlir::TypeRange resultType;
+ auto simdLoopOp = firOpBuilder.create<mlir::omp::SimdLoopOp>(
+ loc, resultType, lowerBound, upperBound, step, alignedVars,
+ /*alignment_values=*/nullptr, ifClauseOperand, nontemporalVars,
+ orderClauseOperand, simdlenClauseOperand, safelenClauseOperand,
+ /*inclusive=*/firOpBuilder.getUnitAttr());
+ createBodyOfOp<mlir::omp::SimdLoopOp>(simdLoopOp, converter, loc, eval,
+ &loopOpClauseList, iv,
+ /*outer=*/false, &dsp);
+}
+
+static void
+createWsLoop(Fortran::lower::AbstractConverter &converter,
+ 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();
+ DataSharingProcessor dsp(converter, beginClauseList, eval);
+ dsp.processStep1();
+
+ Fortran::lower::StatementContext stmtCtx;
+ mlir::Value scheduleChunkClauseOperand;
+ llvm::SmallVector<mlir::Value> lowerBound, upperBound, step, reductionVars,
+ linearVars, linearStepVars;
+ llvm::SmallVector<const Fortran::semantics::Symbol *> iv;
+ llvm::SmallVector<mlir::Attribute> reductionDeclSymbols;
mlir::omp::ClauseOrderKindAttr orderClauseOperand;
mlir::omp::ClauseScheduleKindAttr scheduleValClauseOperand;
- mlir::omp::ScheduleModifierAttr scheduleModClauseOperand;
mlir::UnitAttr nowaitClauseOperand, scheduleSimdClauseOperand;
- llvm::SmallVector<mlir::Attribute> reductionDeclSymbols;
- Fortran::lower::StatementContext stmtCtx;
+ mlir::IntegerAttr orderedClauseOperand;
+ mlir::omp::ScheduleModifierAttr scheduleModClauseOperand;
std::size_t loopVarTypeSize;
- llvm::SmallVector<const Fortran::semantics::Symbol *> iv;
+ ClauseProcessor cp(converter, beginClauseList);
+ cp.processCollapse(loc, eval, lowerBound, upperBound, step, iv,
+ loopVarTypeSize);
+ cp.processScheduleChunk(stmtCtx, scheduleChunkClauseOperand);
+ cp.processReduction(loc, reductionVars, reductionDeclSymbols);
+ cp.processTODO<Fortran::parser::OmpClause::Linear,
+ Fortran::parser::OmpClause::Order>(loc, ompDirective);
+
+ // The types of lower bound, upper bound, and step are converted into the
+ // type of the loop variable if necessary.
+ mlir::Type loopVarType = getLoopVarType(converter, loopVarTypeSize);
+ for (unsigned it = 0; it < (unsigned)lowerBound.size(); it++) {
+ lowerBound[it] =
+ firOpBuilder.createConvert(loc, loopVarType, lowerBound[it]);
+ upperBound[it] =
+ firOpBuilder.createConvert(loc, loopVarType, upperBound[it]);
+ step[it] = firOpBuilder.createConvert(loc, loopVarType, step[it]);
+ }
+
+ auto wsLoopOp = firOpBuilder.create<mlir::omp::WsLoopOp>(
+ loc, lowerBound, upperBound, step, linearVars, linearStepVars,
+ reductionVars,
+ reductionDeclSymbols.empty()
+ ? nullptr
+ : mlir::ArrayAttr::get(firOpBuilder.getContext(),
+ reductionDeclSymbols),
+ scheduleValClauseOperand, scheduleChunkClauseOperand,
+ /*schedule_modifiers=*/nullptr,
+ /*simd_modifier=*/nullptr, nowaitClauseOperand, orderedClauseOperand,
+ orderClauseOperand,
+ /*inclusive=*/firOpBuilder.getUnitAttr());
+
+ // Handle attribute based clauses.
+ if (cp.processOrdered(orderedClauseOperand))
+ wsLoopOp.setOrderedValAttr(orderedClauseOperand);
+
+ if (cp.processSchedule(scheduleValClauseOperand, scheduleModClauseOperand,
+ scheduleSimdClauseOperand)) {
+ wsLoopOp.setScheduleValAttr(scheduleValClauseOperand);
+ wsLoopOp.setScheduleModifierAttr(scheduleModClauseOperand);
+ wsLoopOp.setSimdModifierAttr(scheduleSimdClauseOperand);
+ }
+ // In FORTRAN `nowait` clause occur at the end of `omp do` directive.
+ // i.e
+ // !$omp do
+ // <...>
+ // !$omp end do nowait
+ if (endClauseList) {
+ if (ClauseProcessor(converter, *endClauseList)
+ .processNowait(nowaitClauseOperand))
+ wsLoopOp.setNowaitAttr(nowaitClauseOperand);
+ }
+
+ createBodyOfOp<mlir::omp::WsLoopOp>(wsLoopOp, converter, loc, eval,
+ &beginClauseList, iv,
+ /*outer=*/false, &dsp);
+}
+
+static void genOMP(Fortran::lower::AbstractConverter &converter,
+ Fortran::lower::pft::Evaluation &eval,
+ Fortran::semantics::SemanticsContext &semanticsContext,
+ const Fortran::parser::OpenMPLoopConstruct &loopConstruct) {
const auto &beginLoopDirective =
std::get<Fortran::parser::OmpBeginLoopDirective>(loopConstruct.t);
const auto &loopOpClauseList =
@@ -2995,6 +3122,17 @@ static void genOMP(Fortran::lower::AbstractConverter &converter,
const auto ompDirective =
std::get<Fortran::parser::OmpLoopDirective>(beginLoopDirective.t).v;
+ const auto *endClauseList = [&]() {
+ using RetTy = const Fortran::parser::OmpClauseList *;
+ if (auto &endLoopDirective =
+ std::get<std::optional<Fortran::parser::OmpEndLoopDirective>>(
+ loopConstruct.t)) {
+ return RetTy(
+ &std::get<Fortran::parser::OmpClauseList>((*endLoopDirective).t));
+ }
+ return RetTy();
+ }();
+
bool validDirective = false;
if (llvm::omp::topTaskloopSet.test(ompDirective)) {
validDirective = true;
@@ -3033,97 +3171,14 @@ static void genOMP(Fortran::lower::AbstractConverter &converter,
")");
}
- DataSharingProcessor dsp(converter, loopOpClauseList, eval);
- dsp.processStep1();
-
- ClauseProcessor cp(converter, loopOpClauseList);
- cp.processCollapse(currentLocation, eval, lowerBound, upperBound, step, iv,
- loopVarTypeSize);
- cp.processScheduleChunk(stmtCtx, scheduleChunkClauseOperand);
- cp.processReduction(currentLocation, reductionVars, reductionDeclSymbols);
- cp.processTODO<Fortran::parser::OmpClause::Linear,
- Fortran::parser::OmpClause::Order>(currentLocation,
- ompDirective);
-
- // The types of lower bound, upper bound, and step are converted into the
- // type of the loop variable if necessary.
- mlir::Type loopVarType = getLoopVarType(converter, loopVarTypeSize);
- for (unsigned it = 0; it < (unsigned)lowerBound.size(); it++) {
- lowerBound[it] = firOpBuilder.createConvert(currentLocation, loopVarType,
- lowerBound[it]);
- upperBound[it] = firOpBuilder.createConvert(currentLocation, loopVarType,
- upperBound[it]);
- step[it] =
- firOpBuilder.createConvert(currentLocation, loopVarType, step[it]);
- }
-
// 2.9.3.1 SIMD construct
if (llvm::omp::allSimdSet.test(ompDirective)) {
- llvm::SmallVector<mlir::Value> alignedVars, nontemporalVars;
- mlir::Value ifClauseOperand;
- mlir::IntegerAttr simdlenClauseOperand, safelenClauseOperand;
- cp.processIf(Fortran::parser::OmpIfClause::DirectiveNameModifier::Simd,
- ifClauseOperand);
- cp.processSimdlen(simdlenClauseOperand);
- cp.processSafelen(safelenClauseOperand);
- cp.processTODO<Fortran::parser::OmpClause::Aligned,
- Fortran::parser::OmpClause::Allocate,
- Fortran::parser::OmpClause::Nontemporal>(currentLocation,
- ompDirective);
-
- mlir::TypeRange resultType;
- auto simdLoopOp = firOpBuilder.create<mlir::omp::SimdLoopOp>(
- currentLocation, resultType, lowerBound, upperBound, step, alignedVars,
- /*alignment_values=*/nullptr, ifClauseOperand, nontemporalVars,
- orderClauseOperand, simdlenClauseOperand, safelenClauseOperand,
- /*inclusive=*/firOpBuilder.getUnitAttr());
- createBodyOfOp<mlir::omp::SimdLoopOp>(
- simdLoopOp, converter, currentLocation, eval, &loopOpClauseList, iv,
- /*outer=*/false, &dsp);
- return;
- }
-
- auto wsLoopOp = firOpBuilder.create<mlir::omp::WsLoopOp>(
- currentLocation, lowerBound, upperBound, step, linearVars, linearStepVars,
- reductionVars,
- reductionDeclSymbols.empty()
- ? nullptr
- : mlir::ArrayAttr::get(firOpBuilder.getContext(),
- reductionDeclSymbols),
- scheduleValClauseOperand, scheduleChunkClauseOperand,
- /*schedule_modifiers=*/nullptr,
- /*simd_modifier=*/nullptr, nowaitClauseOperand, orderedClauseOperand,
- orderClauseOperand,
- /*inclusive=*/firOpBuilder.getUnitAttr());
-
- // Handle attribute based clauses.
- if (cp.processOrdered(orderedClauseOperand))
- wsLoopOp.setOrderedValAttr(orderedClauseOperand);
-
- if (cp.processSchedule(scheduleValClauseOperand, scheduleModClauseOperand,
- scheduleSimdClauseOperand)) {
- wsLoopOp.setScheduleValAttr(scheduleValClauseOperand);
- wsLoopOp.setScheduleModifierAttr(scheduleModClauseOperand);
- wsLoopOp.setSimdModifierAttr(scheduleSimdClauseOperand);
- }
- // In FORTRAN `nowait` clause occur at the end of `omp do` directive.
- // i.e
- // !$omp do
- // <...>
- // !$omp end do nowait
- if (const auto &endClauseList =
- std::get<std::optional<Fortran::parser::OmpEndLoopDirective>>(
- loopConstruct.t)) {
- const auto &clauseList =
- std::get<Fortran::parser::OmpClauseList>((*endClauseList).t);
- if (ClauseProcessor(converter, clauseList)
- .processNowait(nowaitClauseOperand))
- wsLoopOp.setNowaitAttr(nowaitClauseOperand);
+ createSimdLoop(converter, eval, ompDirective, loopOpClauseList,
+ currentLocation);
+ } else {
+ createWsLoop(converter, eval, ompDirective, loopOpClauseList, endClauseList,
+ currentLocation);
}
-
- createBodyOfOp<mlir::omp::WsLoopOp>(wsLoopOp, converter, currentLocation,
- eval, &loopOpClauseList, iv,
- /*outer=*/false, &dsp);
}
static void
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
These two constructs were both handled in `genOMP` for loop constructs. There is some shared code between the two, but there are also enough differences to separate these two cases into individual functions. The shared code may be placed into a helper function later if needed. Recursive lowering [1/5]
dfdee7e
to
62f3165
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looks OK to me. Please wait for @skatrak.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thank you Krzysztof for this refactoring, I agree this is a good improvement to make. I just have a small comment, but otherwise it LGTM.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM! I'd just suggest waiting for the buildbot to complete before merging if you haven't already checked locally that the reordering of the convertLoopBounds
call to after clause processing in createSimdLoop
breaks any check-flang lit unit test. Thanks again!
…llvm#77757) These two constructs were both handled in `genOMP` for loop constructs. There is some shared code between the two, but there are also enough differences to separate these two cases into individual functions. The shared code converting loop bounds and steps has been extracted into a separate function. Recursive lowering [1/5]
These two constructs were both handled in
genOMP
for loop constructs. There is some shared code between the two, but there are also enough differences to separate these two cases into individual functions.The shared code may be placed into a helper function later if needed.
Recursive lowering [1/5]