Skip to content

[flang][MLIR] Support delayed privatization for wsloop (PFT -> MLIR) #118271

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.

Sign up for GitHub

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

Merged
merged 1 commit into from
Dec 2, 2024
Merged

[flang][MLIR] Support delayed privatization for wsloop (PFT -> MLIR) #118271

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 4 additions & 4 deletions flang/lib/Lower/OpenMP/OpenMP.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2042,19 +2042,19 @@ static void genStandaloneDo(lower::AbstractConverter &converter,
genWsloopClauses(converter, semaCtx, stmtCtx, item->clauses, loc,
wsloopClauseOps, wsloopReductionSyms);

// TODO: Support delayed privatization.
DataSharingProcessor dsp(converter, semaCtx, item->clauses, eval,
/*shouldCollectPreDeterminedSymbols=*/true,
/*useDelayedPrivatization=*/false, &symTable);
dsp.processStep1();
enableDelayedPrivatizationStaging, &symTable);
dsp.processStep1(&wsloopClauseOps);

mlir::omp::LoopNestOperands loopNestClauseOps;
llvm::SmallVector<const semantics::Symbol *> iv;
genLoopNestClauses(converter, semaCtx, eval, item->clauses, loc,
loopNestClauseOps, iv);

EntryBlockArgs wsloopArgs;
// TODO: Add private syms and vars.
wsloopArgs.priv.syms = dsp.getDelayedPrivSymbols();
wsloopArgs.priv.vars = wsloopClauseOps.privateVars;
wsloopArgs.reduction.syms = wsloopReductionSyms;
wsloopArgs.reduction.vars = wsloopClauseOps.reductionVars;
auto wsloopOp = genWrapperOp<mlir::omp::WsloopOp>(
Expand Down
42 changes: 42 additions & 0 deletions flang/test/Lower/OpenMP/DelayedPrivatization/wsloop.f90
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,42 @@
! RUN: %flang_fc1 -emit-hlfir -fopenmp -mmlir --openmp-enable-delayed-privatization-staging \
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

There may be an obvious thing I am missing here, but is there a reason for "staging" here? Last I did some tests on delayed privatisation, the flag was "openmp-enable-delayed-privatization", which sounded better.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

There are 2 flags. The staging one is for partial implementations of delayed privatization (i.e. where the relevant construct is not support across the entire pipeline; e.g. the MLIR to LLVM lowering is still missing like this case). The one you mentioned is for completely implemented constructs (like omp.parallel).

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks. Understood. LGTM in that case.

! RUN: -o - %s 2>&1 | FileCheck %s
! RUN: bbc -emit-hlfir -fopenmp --openmp-enable-delayed-privatization-staging -o - %s 2>&1 \
! RUN: | FileCheck %s

subroutine wsloop_private
implicit none
integer :: x, i

!$omp parallel do firstprivate(x)
do i = 0, 10
x = x + i
end do
end subroutine wsloop_private

! CHECK: omp.private {type = private} @[[I_PRIVATIZER:.*i_private_ref_i32]]
! CHECK: omp.private {type = firstprivate} @[[X_PRIVATIZER:.*x_firstprivate_ref_i32]]

! CHECK: func.func @{{.*}}() {
! CHECK: %[[I_DECL:.*]]:2 = hlfir.declare %{{.*}} {uniq_name = "{{.*}}i"}
! CHECK: %[[X_DECL:.*]]:2 = hlfir.declare %{{.*}} {uniq_name = "{{.*}}x"}

! CHECK: omp.parallel {
! CHECK: omp.wsloop private(
! CHECK-SAME: @[[X_PRIVATIZER]] %[[X_DECL]]#0 -> %[[X_ARG:[^[:space:]]+]],
! CHECK-SAME: @[[I_PRIVATIZER]] %[[I_DECL]]#0 -> %[[I_ARG:.*]] : {{.*}}) {

! CHECK: omp.loop_nest (%[[IV:.*]]) : i32 = {{.*}} {
! CHECK: %[[X_PRIV_DECL:.*]]:2 = hlfir.declare %[[X_ARG]] {uniq_name = "{{.*}}x"}
! CHECK: %[[I_PRIV_DECL:.*]]:2 = hlfir.declare %[[I_ARG]] {uniq_name = "{{.*}}i"}
! CHECK: fir.store %[[IV]] to %[[I_PRIV_DECL]]#1
! CHECK: %[[X_VAL:.*]] = fir.load %[[X_PRIV_DECL]]#0
! CHECK: %[[I_VAL:.*]] = fir.load %[[I_PRIV_DECL]]#0
! CHECK: %[[ADD_VAL:.*]] = arith.addi %[[X_VAL]], %[[I_VAL]]
! CHECK: hlfir.assign %[[ADD_VAL]] to %[[X_PRIV_DECL]]#0
! CHECK: omp.yield
! CHECK: }
! CHECK: }

! CHECK: omp.terminator
! CHECK: }
! CHECK: }
4 changes: 2 additions & 2 deletions mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2066,8 +2066,8 @@ void WsloopOp::build(OpBuilder &builder, OperationState &state,
builder, state,
/*allocate_vars=*/{}, /*allocator_vars=*/{}, clauses.linearVars,
clauses.linearStepVars, clauses.nowait, clauses.order, clauses.orderMod,
clauses.ordered, /*private_vars=*/{}, /*private_syms=*/nullptr,
clauses.reductionVars,
clauses.ordered, clauses.privateVars,
makeArrayAttr(ctx, clauses.privateSyms), clauses.reductionVars,
makeDenseBoolArrayAttr(ctx, clauses.reductionByref),
makeArrayAttr(ctx, clauses.reductionSyms), clauses.scheduleKind,
clauses.scheduleChunk, clauses.scheduleMod, clauses.scheduleSimd);
Expand Down