Skip to content

[OpenACC][CIR] Lower 'wait' clause for compute/data constructs #137359

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
Apr 25, 2025
Merged

[OpenACC][CIR] Lower 'wait' clause for compute/data constructs #137359

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
3 changes: 3 additions & 0 deletions clang/include/clang/AST/OpenACCClause.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -565,6 +565,9 @@ class OpenACCWaitClause final
llvm::ArrayRef<Expr *> getQueueIdExprs() const {
return OpenACCClauseWithExprs::getExprs().drop_front();
}
// If this is a plain `wait` (no parens) this returns 'false'. Else Sema/Parse
// ensures we have at least one QueueId expression.
bool hasExprs() const { return getLParenLoc().isValid(); }
};

class OpenACCNumGangsClause final
Expand Down
2 changes: 1 addition & 1 deletion clang/lib/AST/OpenACCClause.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -865,7 +865,7 @@ void OpenACCClausePrinter::VisitReductionClause(

void OpenACCClausePrinter::VisitWaitClause(const OpenACCWaitClause &C) {
OS << "wait";
if (!C.getLParenLoc().isInvalid()) {
if (C.hasExprs()) {
OS << "(";
if (C.hasDevNumExpr()) {
OS << "devnum: ";
Expand Down
47 changes: 47 additions & 0 deletions clang/lib/CIR/CodeGen/CIRGenStmtOpenACC.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -400,6 +400,53 @@ class OpenACCClauseCIREmitter final
}
}

void VisitWaitClause(const OpenACCWaitClause &clause) {
if constexpr (isOneOfTypes<OpTy, ParallelOp, SerialOp, KernelsOp, DataOp>) {
if (!clause.hasExprs()) {
operation.setWaitOnlyAttr(
handleDeviceTypeAffectedClause(operation.getWaitOnlyAttr()));
} else {
llvm::SmallVector<mlir::Value> values;

if (clause.hasDevNumExpr())
values.push_back(createIntExpr(clause.getDevNumExpr()));
for (const Expr *E : clause.getQueueIdExprs())
values.push_back(createIntExpr(E));

llvm::SmallVector<int32_t> segments;
if (operation.getWaitOperandsSegments())
llvm::copy(*operation.getWaitOperandsSegments(),
std::back_inserter(segments));

unsigned beforeSegmentSize = segments.size();

mlir::MutableOperandRange range = operation.getWaitOperandsMutable();
operation.setWaitOperandsDeviceTypeAttr(handleDeviceTypeAffectedClause(
operation.getWaitOperandsDeviceTypeAttr(), values, range,
segments));
operation.setWaitOperandsSegments(segments);

// In addition to having to set the 'segments', wait also has a list of
// bool attributes whether it is annotated with 'devnum'. We can use
// our knowledge of how much the 'segments' array grew to determine how
// many we need to add.
llvm::SmallVector<bool> hasDevNums;
if (operation.getHasWaitDevnumAttr())
for (mlir::Attribute A : operation.getHasWaitDevnumAttr())
hasDevNums.push_back(cast<mlir::BoolAttr>(A).getValue());

hasDevNums.insert(hasDevNums.end(), segments.size() - beforeSegmentSize,
clause.hasDevNumExpr());

operation.setHasWaitDevnumAttr(builder.getBoolArrayAttr(hasDevNums));
}
} else {
// TODO: When we've implemented this for everything, switch this to an
// unreachable. Enter data, exit data, update, Combined constructs remain.
return clauseNotImplemented(clause);
}
}

void VisitDefaultAsyncClause(const OpenACCDefaultAsyncClause &clause) {
if constexpr (isOneOfTypes<OpTy, SetOp>) {
operation.getDefaultAsyncMutable().append(
Expand Down
2 changes: 1 addition & 1 deletion clang/lib/Sema/TreeTransform.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12228,7 +12228,7 @@ void OpenACCClauseTransform<Derived>::VisitVectorClause(
template <typename Derived>
void OpenACCClauseTransform<Derived>::VisitWaitClause(
const OpenACCWaitClause &C) {
if (!C.getLParenLoc().isInvalid()) {
if (C.hasExprs()) {
Expr *DevNumExpr = nullptr;
llvm::SmallVector<Expr *> InstantiatedQueueIdExprs;

Expand Down
112 changes: 112 additions & 0 deletions clang/test/CIR/CodeGenOpenACC/data.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -109,5 +109,117 @@ void acc_data(int cond) {
// CHECK-NEXT: acc.terminator
// CHECK-NEXT: } attributes {defaultAttr = #acc<defaultvalue none>}

#pragma acc data default(none) wait
{}
// CHECK-NEXT: acc.data wait {
// CHECK-NEXT: acc.terminator
// CHECK-NEXT: attributes {defaultAttr = #acc<defaultvalue none>}

#pragma acc data default(none) wait device_type(nvidia) wait
{}
// CHECK-NEXT: acc.data wait([#acc.device_type<none>, #acc.device_type<nvidia>]) {
// CHECK-NEXT: acc.terminator
// CHECK-NEXT: attributes {defaultAttr = #acc<defaultvalue none>}

#pragma acc data default(none) wait(1) device_type(nvidia) wait
{}
// CHECK-NEXT: %[[ONE_LITERAL:.*]] = cir.const #cir.int<1> : !s32i
// CHECK-NEXT: %[[ONE_CAST:.*]] = builtin.unrealized_conversion_cast %[[ONE_LITERAL]] : !s32i to si32
// CHECK-NEXT: acc.data wait([#acc.device_type<nvidia>], {%[[ONE_CAST]] : si32}) {
// CHECK-NEXT: acc.terminator
// CHECK-NEXT: attributes {defaultAttr = #acc<defaultvalue none>}

#pragma acc data default(none) wait device_type(nvidia) wait(1)
{}
// CHECK-NEXT: %[[ONE_LITERAL:.*]] = cir.const #cir.int<1> : !s32i
// CHECK-NEXT: %[[ONE_CAST:.*]] = builtin.unrealized_conversion_cast %[[ONE_LITERAL]] : !s32i to si32
// CHECK-NEXT: acc.data wait([#acc.device_type<none>], {%[[ONE_CAST]] : si32} [#acc.device_type<nvidia>]) {
// CHECK-NEXT: acc.terminator
// CHECK-NEXT: attributes {defaultAttr = #acc<defaultvalue none>}

#pragma acc data default(none) wait(1) device_type(nvidia) wait(1)
{}
// CHECK-NEXT: %[[ONE_LITERAL:.*]] = cir.const #cir.int<1> : !s32i
// CHECK-NEXT: %[[ONE_CAST:.*]] = builtin.unrealized_conversion_cast %[[ONE_LITERAL]] : !s32i to si32
// CHECK-NEXT: %[[ONE_LITERAL2:.*]] = cir.const #cir.int<1> : !s32i
// CHECK-NEXT: %[[ONE_CAST2:.*]] = builtin.unrealized_conversion_cast %[[ONE_LITERAL2]] : !s32i to si32
// CHECK-NEXT: acc.data wait({%[[ONE_CAST]] : si32}, {%[[ONE_CAST2]] : si32} [#acc.device_type<nvidia>]) {
// CHECK-NEXT: acc.terminator
// CHECK-NEXT: attributes {defaultAttr = #acc<defaultvalue none>}

#pragma acc data default(none) wait(devnum: cond : 1)
{}
// CHECK-NEXT: %[[COND_LOAD:.*]] = cir.load %[[COND]] : !cir.ptr<!s32i>, !s32i
// CHECK-NEXT: %[[CONV_CAST:.*]] = builtin.unrealized_conversion_cast %[[COND_LOAD]] : !s32i to si32
// CHECK-NEXT: %[[ONE_LITERAL:.*]] = cir.const #cir.int<1> : !s32i
// CHECK-NEXT: %[[ONE_CAST:.*]] = builtin.unrealized_conversion_cast %[[ONE_LITERAL]] : !s32i to si32
// CHECK-NEXT: acc.data wait({devnum: %[[CONV_CAST]] : si32, %[[ONE_CAST]] : si32}) {
// CHECK-NEXT: acc.terminator
// CHECK-NEXT: attributes {defaultAttr = #acc<defaultvalue none>}

#pragma acc data default(none) wait(devnum: cond : 1) device_type(nvidia) wait(devnum: cond : 1)
{}
// CHECK-NEXT: %[[COND_LOAD:.*]] = cir.load %[[COND]] : !cir.ptr<!s32i>, !s32i
// CHECK-NEXT: %[[CONV_CAST:.*]] = builtin.unrealized_conversion_cast %[[COND_LOAD]] : !s32i to si32
// CHECK-NEXT: %[[ONE_LITERAL:.*]] = cir.const #cir.int<1> : !s32i
// CHECK-NEXT: %[[ONE_CAST:.*]] = builtin.unrealized_conversion_cast %[[ONE_LITERAL]] : !s32i to si32
// CHECK-NEXT: %[[COND_LOAD:.*]] = cir.load %[[COND]] : !cir.ptr<!s32i>, !s32i
// CHECK-NEXT: %[[CONV_CAST2:.*]] = builtin.unrealized_conversion_cast %[[COND_LOAD]] : !s32i to si32
// CHECK-NEXT: %[[ONE_LITERAL:.*]] = cir.const #cir.int<1> : !s32i
// CHECK-NEXT: %[[ONE_CAST2:.*]] = builtin.unrealized_conversion_cast %[[ONE_LITERAL]] : !s32i to si32
// CHECK-NEXT: acc.data wait({devnum: %[[CONV_CAST]] : si32, %[[ONE_CAST]] : si32}, {devnum: %[[CONV_CAST2]] : si32, %[[ONE_CAST2]] : si32} [#acc.device_type<nvidia>]) {
// CHECK-NEXT: acc.terminator
// CHECK-NEXT: attributes {defaultAttr = #acc<defaultvalue none>}

#pragma acc data default(none) wait(devnum: cond : 1, 2)
{}
// CHECK-NEXT: %[[COND_LOAD:.*]] = cir.load %[[COND]] : !cir.ptr<!s32i>, !s32i
// CHECK-NEXT: %[[CONV_CAST:.*]] = builtin.unrealized_conversion_cast %[[COND_LOAD]] : !s32i to si32
// CHECK-NEXT: %[[ONE_LITERAL:.*]] = cir.const #cir.int<1> : !s32i
// CHECK-NEXT: %[[ONE_CAST:.*]] = builtin.unrealized_conversion_cast %[[ONE_LITERAL]] : !s32i to si32
// CHECK-NEXT: %[[TWO_LITERAL:.*]] = cir.const #cir.int<2> : !s32i
// CHECK-NEXT: %[[TWO_CAST:.*]] = builtin.unrealized_conversion_cast %[[TWO_LITERAL]] : !s32i to si32
// CHECK-NEXT: acc.data wait({devnum: %[[CONV_CAST]] : si32, %[[ONE_CAST]] : si32, %[[TWO_CAST]] : si32}) {
// CHECK-NEXT: acc.terminator
// CHECK-NEXT: attributes {defaultAttr = #acc<defaultvalue none>}

#pragma acc data default(none) wait(devnum: cond : 1, 2) device_type(nvidia, radeon) wait(devnum: cond : 1, 2)
{}
// CHECK-NEXT: %[[COND_LOAD:.*]] = cir.load %[[COND]] : !cir.ptr<!s32i>, !s32i
// CHECK-NEXT: %[[CONV_CAST:.*]] = builtin.unrealized_conversion_cast %[[COND_LOAD]] : !s32i to si32
// CHECK-NEXT: %[[ONE_LITERAL:.*]] = cir.const #cir.int<1> : !s32i
// CHECK-NEXT: %[[ONE_CAST:.*]] = builtin.unrealized_conversion_cast %[[ONE_LITERAL]] : !s32i to si32
// CHECK-NEXT: %[[TWO_LITERAL:.*]] = cir.const #cir.int<2> : !s32i
// CHECK-NEXT: %[[TWO_CAST:.*]] = builtin.unrealized_conversion_cast %[[TWO_LITERAL]] : !s32i to si32
// CHECK-NEXT: %[[COND_LOAD:.*]] = cir.load %[[COND]] : !cir.ptr<!s32i>, !s32i
// CHECK-NEXT: %[[CONV_CAST2:.*]] = builtin.unrealized_conversion_cast %[[COND_LOAD]] : !s32i to si32
// CHECK-NEXT: %[[ONE_LITERAL:.*]] = cir.const #cir.int<1> : !s32i
// CHECK-NEXT: %[[ONE_CAST2:.*]] = builtin.unrealized_conversion_cast %[[ONE_LITERAL]] : !s32i to si32
// CHECK-NEXT: %[[TWO_LITERAL:.*]] = cir.const #cir.int<2> : !s32i
// CHECK-NEXT: %[[TWO_CAST2:.*]] = builtin.unrealized_conversion_cast %[[TWO_LITERAL]] : !s32i to si32
// CHECK-NEXT: acc.data wait({devnum: %[[CONV_CAST]] : si32, %[[ONE_CAST]] : si32, %[[TWO_CAST]] : si32}, {devnum: %[[CONV_CAST2]] : si32, %[[ONE_CAST2]] : si32, %[[TWO_CAST2]] : si32} [#acc.device_type<nvidia>], {devnum: %[[CONV_CAST2]] : si32, %[[ONE_CAST2]] : si32, %[[TWO_CAST2]] : si32} [#acc.device_type<radeon>]) {
// CHECK-NEXT: acc.terminator
// CHECK-NEXT: attributes {defaultAttr = #acc<defaultvalue none>}

#pragma acc data default(none) wait(cond, 1)
{}
// CHECK-NEXT: %[[COND_LOAD:.*]] = cir.load %[[COND]] : !cir.ptr<!s32i>, !s32i
// CHECK-NEXT: %[[CONV_CAST:.*]] = builtin.unrealized_conversion_cast %[[COND_LOAD]] : !s32i to si32
// CHECK-NEXT: %[[ONE_LITERAL:.*]] = cir.const #cir.int<1> : !s32i
// CHECK-NEXT: %[[ONE_CAST:.*]] = builtin.unrealized_conversion_cast %[[ONE_LITERAL]] : !s32i to si32
// CHECK-NEXT: acc.data wait({%[[CONV_CAST]] : si32, %[[ONE_CAST]] : si32}) {
// CHECK-NEXT: acc.terminator
// CHECK-NEXT: attributes {defaultAttr = #acc<defaultvalue none>}

#pragma acc data default(none) wait(queues: cond, 1) device_type(radeon)
{}
// CHECK-NEXT: %[[COND_LOAD:.*]] = cir.load %[[COND]] : !cir.ptr<!s32i>, !s32i
// CHECK-NEXT: %[[CONV_CAST:.*]] = builtin.unrealized_conversion_cast %[[COND_LOAD]] : !s32i to si32
// CHECK-NEXT: %[[ONE_LITERAL:.*]] = cir.const #cir.int<1> : !s32i
// CHECK-NEXT: %[[ONE_CAST:.*]] = builtin.unrealized_conversion_cast %[[ONE_LITERAL]] : !s32i to si32
// CHECK-NEXT: acc.data wait({%[[CONV_CAST]] : si32, %[[ONE_CAST]] : si32}) {
// CHECK-NEXT: acc.terminator
// CHECK-NEXT: attributes {defaultAttr = #acc<defaultvalue none>}

// CHECK-NEXT: cir.return
}
112 changes: 112 additions & 0 deletions clang/test/CIR/CodeGenOpenACC/kernels.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -302,5 +302,117 @@ void acc_kernels(int cond) {
// CHECK-NEXT: acc.terminator
// CHECK-NEXT: } loc

#pragma acc kernels wait
{}
// CHECK-NEXT: acc.kernels wait {
// CHECK-NEXT: acc.terminator
// CHECK-NEXT: } loc

#pragma acc kernels wait device_type(nvidia) wait
{}
// CHECK-NEXT: acc.kernels wait([#acc.device_type<none>, #acc.device_type<nvidia>]) {
// CHECK-NEXT: acc.terminator
// CHECK-NEXT: } loc

#pragma acc kernels wait(1) device_type(nvidia) wait
{}
// CHECK-NEXT: %[[ONE_LITERAL:.*]] = cir.const #cir.int<1> : !s32i
// CHECK-NEXT: %[[ONE_CAST:.*]] = builtin.unrealized_conversion_cast %[[ONE_LITERAL]] : !s32i to si32
// CHECK-NEXT: acc.kernels wait([#acc.device_type<nvidia>], {%[[ONE_CAST]] : si32}) {
// CHECK-NEXT: acc.terminator
// CHECK-NEXT: } loc

#pragma acc kernels wait device_type(nvidia) wait(1)
{}
// CHECK-NEXT: %[[ONE_LITERAL:.*]] = cir.const #cir.int<1> : !s32i
// CHECK-NEXT: %[[ONE_CAST:.*]] = builtin.unrealized_conversion_cast %[[ONE_LITERAL]] : !s32i to si32
// CHECK-NEXT: acc.kernels wait([#acc.device_type<none>], {%[[ONE_CAST]] : si32} [#acc.device_type<nvidia>]) {
// CHECK-NEXT: acc.terminator
// CHECK-NEXT: } loc

#pragma acc kernels wait(1) device_type(nvidia) wait(1)
{}
// CHECK-NEXT: %[[ONE_LITERAL:.*]] = cir.const #cir.int<1> : !s32i
// CHECK-NEXT: %[[ONE_CAST:.*]] = builtin.unrealized_conversion_cast %[[ONE_LITERAL]] : !s32i to si32
// CHECK-NEXT: %[[ONE_LITERAL2:.*]] = cir.const #cir.int<1> : !s32i
// CHECK-NEXT: %[[ONE_CAST2:.*]] = builtin.unrealized_conversion_cast %[[ONE_LITERAL2]] : !s32i to si32
// CHECK-NEXT: acc.kernels wait({%[[ONE_CAST]] : si32}, {%[[ONE_CAST2]] : si32} [#acc.device_type<nvidia>]) {
// CHECK-NEXT: acc.terminator
// CHECK-NEXT: } loc

#pragma acc kernels wait(devnum: cond : 1)
{}
// CHECK-NEXT: %[[COND_LOAD:.*]] = cir.load %[[COND]] : !cir.ptr<!s32i>, !s32i
// CHECK-NEXT: %[[CONV_CAST:.*]] = builtin.unrealized_conversion_cast %[[COND_LOAD]] : !s32i to si32
// CHECK-NEXT: %[[ONE_LITERAL:.*]] = cir.const #cir.int<1> : !s32i
// CHECK-NEXT: %[[ONE_CAST:.*]] = builtin.unrealized_conversion_cast %[[ONE_LITERAL]] : !s32i to si32
// CHECK-NEXT: acc.kernels wait({devnum: %[[CONV_CAST]] : si32, %[[ONE_CAST]] : si32}) {
// CHECK-NEXT: acc.terminator
// CHECK-NEXT: } loc

#pragma acc kernels wait(devnum: cond : 1) device_type(nvidia) wait(devnum: cond : 1)
{}
// CHECK-NEXT: %[[COND_LOAD:.*]] = cir.load %[[COND]] : !cir.ptr<!s32i>, !s32i
// CHECK-NEXT: %[[CONV_CAST:.*]] = builtin.unrealized_conversion_cast %[[COND_LOAD]] : !s32i to si32
// CHECK-NEXT: %[[ONE_LITERAL:.*]] = cir.const #cir.int<1> : !s32i
// CHECK-NEXT: %[[ONE_CAST:.*]] = builtin.unrealized_conversion_cast %[[ONE_LITERAL]] : !s32i to si32
// CHECK-NEXT: %[[COND_LOAD:.*]] = cir.load %[[COND]] : !cir.ptr<!s32i>, !s32i
// CHECK-NEXT: %[[CONV_CAST2:.*]] = builtin.unrealized_conversion_cast %[[COND_LOAD]] : !s32i to si32
// CHECK-NEXT: %[[ONE_LITERAL:.*]] = cir.const #cir.int<1> : !s32i
// CHECK-NEXT: %[[ONE_CAST2:.*]] = builtin.unrealized_conversion_cast %[[ONE_LITERAL]] : !s32i to si32
// CHECK-NEXT: acc.kernels wait({devnum: %[[CONV_CAST]] : si32, %[[ONE_CAST]] : si32}, {devnum: %[[CONV_CAST2]] : si32, %[[ONE_CAST2]] : si32} [#acc.device_type<nvidia>]) {
// CHECK-NEXT: acc.terminator
// CHECK-NEXT: } loc

#pragma acc kernels wait(devnum: cond : 1, 2)
{}
// CHECK-NEXT: %[[COND_LOAD:.*]] = cir.load %[[COND]] : !cir.ptr<!s32i>, !s32i
// CHECK-NEXT: %[[CONV_CAST:.*]] = builtin.unrealized_conversion_cast %[[COND_LOAD]] : !s32i to si32
// CHECK-NEXT: %[[ONE_LITERAL:.*]] = cir.const #cir.int<1> : !s32i
// CHECK-NEXT: %[[ONE_CAST:.*]] = builtin.unrealized_conversion_cast %[[ONE_LITERAL]] : !s32i to si32
// CHECK-NEXT: %[[TWO_LITERAL:.*]] = cir.const #cir.int<2> : !s32i
// CHECK-NEXT: %[[TWO_CAST:.*]] = builtin.unrealized_conversion_cast %[[TWO_LITERAL]] : !s32i to si32
// CHECK-NEXT: acc.kernels wait({devnum: %[[CONV_CAST]] : si32, %[[ONE_CAST]] : si32, %[[TWO_CAST]] : si32}) {
// CHECK-NEXT: acc.terminator
// CHECK-NEXT: } loc

#pragma acc kernels wait(devnum: cond : 1, 2) device_type(nvidia, radeon) wait(devnum: cond : 1, 2)
{}
// CHECK-NEXT: %[[COND_LOAD:.*]] = cir.load %[[COND]] : !cir.ptr<!s32i>, !s32i
// CHECK-NEXT: %[[CONV_CAST:.*]] = builtin.unrealized_conversion_cast %[[COND_LOAD]] : !s32i to si32
// CHECK-NEXT: %[[ONE_LITERAL:.*]] = cir.const #cir.int<1> : !s32i
// CHECK-NEXT: %[[ONE_CAST:.*]] = builtin.unrealized_conversion_cast %[[ONE_LITERAL]] : !s32i to si32
// CHECK-NEXT: %[[TWO_LITERAL:.*]] = cir.const #cir.int<2> : !s32i
// CHECK-NEXT: %[[TWO_CAST:.*]] = builtin.unrealized_conversion_cast %[[TWO_LITERAL]] : !s32i to si32
// CHECK-NEXT: %[[COND_LOAD:.*]] = cir.load %[[COND]] : !cir.ptr<!s32i>, !s32i
// CHECK-NEXT: %[[CONV_CAST2:.*]] = builtin.unrealized_conversion_cast %[[COND_LOAD]] : !s32i to si32
// CHECK-NEXT: %[[ONE_LITERAL:.*]] = cir.const #cir.int<1> : !s32i
// CHECK-NEXT: %[[ONE_CAST2:.*]] = builtin.unrealized_conversion_cast %[[ONE_LITERAL]] : !s32i to si32
// CHECK-NEXT: %[[TWO_LITERAL:.*]] = cir.const #cir.int<2> : !s32i
// CHECK-NEXT: %[[TWO_CAST2:.*]] = builtin.unrealized_conversion_cast %[[TWO_LITERAL]] : !s32i to si32
// CHECK-NEXT: acc.kernels wait({devnum: %[[CONV_CAST]] : si32, %[[ONE_CAST]] : si32, %[[TWO_CAST]] : si32}, {devnum: %[[CONV_CAST2]] : si32, %[[ONE_CAST2]] : si32, %[[TWO_CAST2]] : si32} [#acc.device_type<nvidia>], {devnum: %[[CONV_CAST2]] : si32, %[[ONE_CAST2]] : si32, %[[TWO_CAST2]] : si32} [#acc.device_type<radeon>]) {
// CHECK-NEXT: acc.terminator
// CHECK-NEXT: } loc

#pragma acc kernels wait(cond, 1)
{}
// CHECK-NEXT: %[[COND_LOAD:.*]] = cir.load %[[COND]] : !cir.ptr<!s32i>, !s32i
// CHECK-NEXT: %[[CONV_CAST:.*]] = builtin.unrealized_conversion_cast %[[COND_LOAD]] : !s32i to si32
// CHECK-NEXT: %[[ONE_LITERAL:.*]] = cir.const #cir.int<1> : !s32i
// CHECK-NEXT: %[[ONE_CAST:.*]] = builtin.unrealized_conversion_cast %[[ONE_LITERAL]] : !s32i to si32
// CHECK-NEXT: acc.kernels wait({%[[CONV_CAST]] : si32, %[[ONE_CAST]] : si32}) {
// CHECK-NEXT: acc.terminator
// CHECK-NEXT: } loc

#pragma acc kernels wait(queues: cond, 1) device_type(radeon)
{}
// CHECK-NEXT: %[[COND_LOAD:.*]] = cir.load %[[COND]] : !cir.ptr<!s32i>, !s32i
// CHECK-NEXT: %[[CONV_CAST:.*]] = builtin.unrealized_conversion_cast %[[COND_LOAD]] : !s32i to si32
// CHECK-NEXT: %[[ONE_LITERAL:.*]] = cir.const #cir.int<1> : !s32i
// CHECK-NEXT: %[[ONE_CAST:.*]] = builtin.unrealized_conversion_cast %[[ONE_LITERAL]] : !s32i to si32
// CHECK-NEXT: acc.kernels wait({%[[CONV_CAST]] : si32, %[[ONE_CAST]] : si32}) {
// CHECK-NEXT: acc.terminator
// CHECK-NEXT: } loc

// CHECK-NEXT: cir.return
}
Loading
Loading